The problem with the older USB Standards is that everything HAD to pass through the CPU during USB transfers. Something that thankfully was finally abandoned with the introduction of USB 3.0 and 3.1. Which i think is definitely the biggest gain from this standard upgrade, beside the file transfer speed.
This is also the reason why Firewire was superior even to USB2, even when it looked sighly slower at specs. Firewire controllers, just as SCSI before them, were always "smarter" than their USB counterparts. The CPU in the FW ASIC of the card and the peripherals could do more by itself and transfer the data between them and the host by using DMA with almost no CPU intervention. Also, the software stack was smaller, which meant you could do things like implementing Boot Support, Debugging Mode and Target Mode (your computer appears as an storage device to another computer) in firmware in a very compact way. Ofc, since firewire controllers were almost microcomputers by themselves, that means the controller chips for both cards and peripherals were more complex, expensive and less scalable to produce than USB counterparts, which in the end was firewire demise.
Just want to say, USB overhead was well known at the time and even latency was an issue. This was all known and tested extensively by nerds a long time ago. USB didn't really become a non-issue until USB 2.0. I remember very well discussing this ad nauseam on forums back in the day.
It looks like USB 1.0 generates interrupt for CPU every time mouse moves (instead using DMA bypassing CPU entirely). Every CPU handle interrupts and context switching with different penalty in cycles lost. Modern CPU has special HW to accelerate context switching due to those were designed for modern multitasking OS. Obviously old CPU designed for single tasking DOS lacks this support.
@@bitsundboltsThis was going to be my suggestion also. Various USB chipsets may show different results. On some more expensive USB ISA/PCI cards each port might get its own controller. Also drivers in various OS's could be a factor. Also, did you see any difference with a wired USB device and wireless? 🙂
I did not test with a wired USB keyboard this time, but in my first video, there was no difference between a wired and a wireless keyboard. I would assume they behave the same.
People might be wondering why soundcards had all those extra ports and not the video cards or something else. Well, in the early days, soundcards were primarily used for gaming,. There wasn't much else that would use sound card, we didn't have music like we do today. It was mostly noise, triangle waves, and such. Multimedia didn't really come to computers until the early 90s. Computers like the Amiga were popular for making music, but it was still quite challenging to store CD-quality audio tracks on your system. Back then, we had about 40-80MB of HDD space, if I remember correctly. Companies knew that if someone bought a soundcard, it was likely to play games. Great video again! 👍
Thank you! And interesting information which makes a lot of sense! Funny enough, I remember most of the soundcards I had back when onboard audio wasn't really a thing. The first soundcard ingot was an ESS688 AudioDrive - it was such an essential piece of Hardware in my AMD 486-DX4-100.
Aside from puny HDD sizes, Amiga's sound chip (Paula) from 1984 was never updated and the it was stuck with its 8-bit DACs till the end. That made it unfit for CD-level quality audio. Generally, the platform was stagnant. Most people bought cheaper 'small Amigas' that lacked Zorro slots to put expansion cards and have more PC-like market for graphic cards (the so-called ReTargettabble Grpahics or RTC - ability to use graphic cards appeared very late, only for 'big' Amigas with expansion slots and - for obvious - reasons only for well-behaved Workbench GUI apps, not for games). To make matters worse, games bypassed the OS and talked to hardware directly for performance, so most targeted the most popular config: a sub 1 MB Amiga 500. That made Amiga a hostage of her (rapidly aging) original chipset: updates were minor to not break compat and due to Commodore skimping on R&D.
13:14 I think I had this exact A4tech serial mouse. Thanks for bringing back the memories! As for the USB performance drop, try external card with a NEC chip. I think I read somewhere that it has lower overhead in comparison to USB cards on a VIA chip.
I'm now installing a serial port on my Ryzen 7 5700X. Might be a tad slower for transferring videos from the iPhone. Next video is then scheduled for 2027! Thanks for the comprehensive analysis!
Retro PCs Tony, Retro PCs :) Please do not postpone your content until 2027! It would be a shame to deprive your community from your great content for so long!
I don't know if it's just my modern perspective of CPUs and USB protocol, but I don't quite understand how a slight performance dip of 1-2% on an older system can make a difference in the long run. But this is excellent information to help me learn how it can have an effect, and I'm glad you went through the effort of testing this. Your videos have been amazing to watch through, from your 3DFX GPU mods to the AMD K6-2+ mod that got me started with your channel. Please, keep working hard on this content, because I hope to learn more in the future.
Something to consider is that since a lot of the games at the time were targeting sub-30 FPS and that most games had their speed directly corresponding to how fast your CPU was (meaning that a faster CPU literally translated into faster game logic, thus the emulation problem were games were running "too fast"). 1-2% difference in performance is certainly meaningful. For a modern comparison, a 2% difference in performance from a game running at 60 FPS could literally translate into a loss of 1.2 FPS which can wreak havoc if you're trying to achieve a 60Hz VSync frame lock. Imaging trying to speedrun a game and then finding out that the mouse and keyboard you have connected changes how fast your car can literally go, or how fast your character can literally run, etc. Sure it's not "insane" performance drops but it's certainly enough to look into (especially considering that we may see even more issues when using expansion cards or moving to USB 2.0, but hopefully BaB can give us that video in the near future to see for ourselves)
I think a much more real world problem is CPU load during USB bulk transfer. This affects any application that requires CPU speed and fast data transfer at the same time. When editing video on a Mac in ~2005 on an external HDD it was strongly recommended to use FireWire instead of USB because FireWire used DMA and caused much less CPU load. The difference could be dramatic. Of course, a few years later faster processors have made that problem mostly irrelevant.
Maybe not irrelevant - because developers have used the extra processing power to implement more complex real time effects in various editing suites, while the need for latency-free recording is as important as ever.
Firewire was nice but e-SATA was better. RIP e-SATA though, when USB 3.1 came out it pretty much made e-SATA pointless. I wish we could have kept e-SATA honestly.
I've added several USB 2.0 cards to my retro rigs but only for quick file transfer. Good to see that I made the right choice keeping my input devices analog. Interesting thing about adding usb cards is the PCI version. You can't just add any generic card because chances are that it uses a higher PCI version than is supported on your old motherboard. There's a pretty good thread over on Vogons about which chipsets to pick and avoid.
The performance drop is waaay bigger than 4%, I remember back in the day I had a PIII 550 Mhz, and when I switched from a PCI 56K to an USB 512 Kbps modem, when playing Counter-Strike, my FPS went down from 50+ to low 30's, it was immediately noticeable, I had to keep it unplugged all the time to just play single player games, I begged my parents for months to switch to another ISP, when they finally got an RJ-45 ethernet gateway it was such a relief, I got back on good form again. Even today in modern 8 core systems, you can clearly see in task manager that when moving a 1000 Hz "gaming mouse" prevents the CPU from getting into the lowest power and clock states, my i7 4770 can downclock to 1 Ghz when moving a PS/2 at 200 "Hz" (reports per second) or 125 Mhz USB mouse, but when having a 1000 Hz + USB microphone, It idles at around 1.3-1.5 Ghz... that is the reason even today, laptop touchpads and keyboards still use the PS/2 protocol, to save CPU power.
It would be cool to re-run the tests with the USB devices actively in use and not just idling, like using the USB soundcard from your headset + KB and mouse moving, there are other things that also affect CPU performance on retro (and modern...) PC's, using the Realtek onboard soundcard (specially if you enable the background noise suppression), and high amounts of open TCP/IP connections, write caching on HDD's, having Hyper Threading enabled when not taking advantage of the extra threads, and some more I don't remember....
Very interesting video! A few details and thoughts: - From the PC's perspective it makes no difference if the USB supports "real" interrupts or not: The USB host controller manages the USB protocol side (including polling the devices) and sends a (real) interrupt to the PC whenever necessary. Some controllers might send unnecessary interrupts and let the driver figure that out, in which case you would have a significantly higher interrupt load. I'm not aware, however, how frequent that behavior is in practice, so let's keep that in the "theories" drawer. On the other hand the fact that on some platforms you tested there were periodic dips in the performance, presumably without any mouse movement, would support that hypothesis. - The problem presumably comes from the fact that the USB driver stack is interrupting the current program and affects the contents of the cache - just as the PS/2 interrupt would - but is probably more complex, accesses more data and affects the cache more than legacy input device drivers. - The specific driver implementation is another issue: If not in legacy emulation mode, the USB communication goes through the driver the manufacturer of the host controller provided. I suspect the implementations will differ wildly. It would be interesting to see how this issue affects Linux, which I suspect might have more consistent and better quality drivers overall. - It would also be interesting to compare different host controller cards, and specifically USB 2.0 ones. - As was mentioned in another comment, legacy emulation could lead to poor performance as well and needs to be tested and compared on a motherboard level basis to understand the impact.
The biggest factor making a difference is the USB controller being used. Some are more heavy in interrupting the CPU than others, who might to more things in hardware before interrupting software.
I am not surprised, this was one of the overclocker arguments from Kingpin when making the EVGA's Socket 1366 motherboards and offering IDE connector for the harddrive and PS/2 for mouse and keyboard. Basically simplify and reduce parasitic loads for benchmarking runs and get that higher score. Of course if you run a less powerful CPU, the performance costs start to become noticeable.
Interesting that it's up to Socket 1366 for this to be an issue. I guess if you're chasing world records, you want to have every little extra performance you can get!
USB penalty in retro devices isn't just hardware issues, but also software issues. The USB stack modules included in BIOSes for KB/Mouse support in DOS mode aren't smaller things and can consume precious UMB blocks you would need for other stuff. Sometimes it will prevent EMS memory from working as the USB BIOS module will steal the EMS window memory block (some Systemsoft BIOSes show the problem, it can be fixed assuming the board manufacturer provided a fixed update for that). This enables SMM/NMI which can conflict with DOS software and DOS drivers which also make use of NMI (sound card drivers or emulators for example). In Windows, USB will always enable WDM (NTKern). Depending in your board and devices, it can conflict with VxD drivers, mostly sound cards. It will consume more memory, and make the OS slower in some circustances.
Amazing video as always! I've appreciated same behaviour as you since you published the previous video showing this, when tested it on a 166MMX machine. But at least for me, having the possibility of having my childhood pc now connected to a big flat tv and wireless keyboard&mouse clearly compensates the 1% penalty! Haha :)
Haha, yeah, it's not that big of a deal. But still weird that we can see a cache performance issue when a USB input device is connected. Nice to hear that you tested this and the behavior is the same for you. Give a bit more credibility to my videos 😉
It was known 20 years ago that USB was/is CPU bound. One of the flame war arguments back then was that firewire was not...... USB was soo convenient that the 1 to 2% performance loss was acceptable to the majority of users. Nowadays it is hard to find even PS2 new keyboards/mice. I just bought a no-name PS2 keyboard from Amazon for an Amiga motherboard with DB9 (Amiga) and PS2 ports, but no USB, I also bought a little USB controller chip which sits on the CIA chips to provide USB keyboard support. AmiKey1200. Cool video. Never before had I seen the effect of USB being CPU bound.
This video was timed perfectly - I may end up needing to use my mouse in USB mode on a baby AT board that I recently purchased as it is missing the PS/2 connector from its LPT bracket, and the brackets from my other ("modern") baby AT board uses a different connector. The system will be running a Celeron 333 (or a Pentium II 233/350) under Windows 3.1 or 95 so quite relevant knowledge.
Glad to hear that the video may give you some insights. Maybe you can run SpeedSys to verify the behavior of the USB device and if it affects the performance.
@@bitsundbolts Considering the performance of a Celeron (with cache) is near on par with a same-clock Pentium II, and with the board (PCPartner VIB862D w/SB Vibra16XV) having the Via VIA VT82C691 and VIA VT82C586B, the results should be accurately comparable. I could test with a Celeron 300A/450A (contact B21 cut) and PIII 650 if you want - both on my new board (since I can select the CPU speed, and I fully expect relatively low performance) and Intel SE440BX-2 (currently setup next to me). Can't reliably test my current PII 350 as its heatsink is too large and it barely fits my boards, and the PII 233 only posts into BIOS (and only at 133MHz with the BIOS jumper set to "config"). I also have a Celeron 500 and 433 that could be tested in a slotket - they work flawlessly on the Intel board.
If you find the time, it would be interesting to see if you can replicate the behavior in any of your retro setups. I do have a few Celeron CPUs here as well, but I haven't tested those yet. I can imagine that the USB issue is present in almost all builds from that time up to around 700/800 MHz
I'll have a go, realistically I shouldn't need to do more than disable the memory manager to get real mode working for the tests. Might also give my mobile Duron 1300 a go on my GA-7IXEH, as it only runs at 500MHz on it and the board has the as far as I can tell unique to it Via KT133E. I'm not going to use my Jetway994 as I still haven't gotten it reliable enough to comfortably say any result is accurate (I've had back-to-back CPU benchmarks vary by over 25% with a barebones setup and only PS/2 mouse and keyboard - and it is unstable as well).
@@bitsundbolts Starting the tests now with the PIII 650, commented out everything in autoexec+dosstart (also copied then cleared config.sys) and started the tests. I just need to remember to enable legacy USB support in the BIOS (I've had no need for it due to using an adapted MS Wheel Mouse Optical and a Fellowes adapted AT keyboard with my analogue KVM).
Nice findings! Please, next time pay the attention if USB Interrupt and/or USB Legacy Support are ON/OFF in the BIOS setup. This may also do a measurable impact on the results.
This! I was thinking watching this video if "Assign IRQ For USB" and "Legacy USB Support" options in the BIOS make a difference. I stopped using USB mice or keyboards on my 90s PCs years ago since it affected the sound output, I could clearly hear interference when using them, no matter what.
Turns out I had the USB vs PS2 inner workings wrong but with more or less the same result XD. Someone told me that PS2 was better because it sends interrupts when it needs to, so you get "no latency" on games. , but USB sends a request and joins the scheduler of the cpu so your input can be "delayed". Your explanation is clearly the correct one. Probably someone noticed a difference when gaming but didn't made the proper tests to see a performance drop.
Interesting observations. I've been using wireless USB input devices on an old P233 MMX system as it is a lot more convenient than using an old crusty DIN keyboard and PS/2 mouse. At least the performance degradation isn't too severe, so the added convenience of USB might be worth the slight performance hit.
Yes, the performance hit is small and won't make a difference if some application or game is already running poorly. Convenience is a very good argument, especially when I don't have to have all those cables around!
It's interesting that you reported this. I have a VIA USB 2.0 PCI card in my PIII-933 system. I use it routinely for large file transfers as I have reliability issues with my optical drives. I also don't expose my current systems to my retro systems over the same network. Honestly, it doesn't affect performance for me to notice it under Windows 98 SE (USB storage extension) and 2000 Pro.
I have not seen any issue with data transfer (HDD or flash drives). The issue only appears when using input devices like keyboards and mouse. I will get some PCI USB 2.0 expansion cards with different chipsets (NEC, ALI, VIA, etc.) and see what is going on there.
I think the reason you are getting cache performance degradation is the bios regularly trapping into SMM for legacy usb support via a timer SMI, and SMM code running from a different location, accessing different data will surely trash your cache.
You may be right, unfortunately, I do not have the technical knowledge. All I can say is that this behavior was present in two socket 7 boards and one slot 1 board. The other slot 1 (P3B-F + PIII 1000) did not show such behavior which may have been due to the fast CPU, but I have to test this board with a slower CPU once more.
I think you mean a USB-to-PS/2 connector. I would assume it to behave like a regular PS/2 device, but I have not tested this yet. It would turn a "bad" USB keyboard into a "good" PS/2 keyboard.
@@bitsundboltspassive adapters require the mouse or keyboard to support ps/2 protocols. I use microsoft wmo 1.1a with passive usb to ps/2 converter. At the bottom of the mouse “usb and ps/2 compatible” is written. Meaning it would work just like a native ps/2 device.
Oh yes, I remember being stuck because of this! You can't just turn any USB device into a PS/2 device. It must be compatible and support the protocol (when using passive adapters)! Thanks for reminding me 😀
@@bitsundbolts exactly! The requirement for being campatible with the protocoll is also a good hint. But still I think using an adapter could help in widening the available peripherals i.e. the standard but "good" cherry keyboards.
Great research! Would be interesting to get an overview of percentages performance loss across core frequencies, especially with the P2 and P3 era. Wondering if it would be linear with clock speed, and at what point it's no longer measureable.
With each video I make about this topic, I have more questions afterwards once I read the comments. Although I have a feeling that around 800 MHz, it will be difficult to measure a difference, it could also be that I have to change benchmarks and testing methodology. For now, it's great to see the graphs in SpeedSys of the cash - at least we know something is there.
I believe results can vary based on controller chip or other factors, but particularly with add in USB2 cards, that the USB controllers being enabled (no device plugged in) in device manager can at times be enough to cause a performance degradation. Of course that means you can just toggle the controller on and off as needed if max performance is desired. I have gotten mixed messages from others on the exact conditions on when the degradation happens, but I know what works for my system. Plenty of fun stuff to test there since USB 1.1 comes in OHCI and UHCI flavors, and USB 2 cards expose one of those on top of the EHCI controller which can be disabled individually. If my memory serves they also as a result, if all controllers are enabled, use 3 of the 4 PCI IRQs, so while not a huge deal IRQ sharing could be a factor in achieving the last bit of performance.
I still use a PS/2 keyboard with my modern system. I would use a PS/2 mouse as well if my motherboard had a second PS/2 port. It only has one PS/2 for keyboard OR mouse. PS/2 ports have 6 pins but only 4 pins are used. On motherboards with only one PS/2 port, the 2 unused pins should be the clk and data for a second PS/2 port via a Y adaptor. All the super I/O chips have 2 PS/2 ports so there is no reason why the clk and data for the second port couldn't be routed to the 2 unused pins on motherboards with only one physical PS/2 connector.
Just a few days ago i was testing an old system equipped with a Pentium 3 667MHz on an AOpen AX34 Pro, 256 MB of SDR PC133 and a Voodoo 3 3000 running windows 98SE. That system has only 2 USB ports in the rear I/O panel so i added a VIA based pci usb card. With a usb stick connected to that card, running 3D mark 99 Max i was getting a final score that was half the result i previously managed to get...unplugged the usb stick and scores were back to normal... Pretty extreme behaviour, i know that VIA pci usb cards are doomed but this behaviour is fare beyond my expectations...
The older connectors are interrupt based, meaning the CPU only needs to do something if there was an input. USB needs to be polled regularly at the rate the USB device tells the host to be in the endpoint descriptors. That the polling rate can have a negative impact isn't just an old problem. Just four months ago Microsoft released an update for high polling rate issues on Win 11. As far as I know there were a lot of issues too back then when the first 1 kHz mouses came out. So not even up to date systems can really properly handle high polling rate devices, so no wonder that on older system can't deal with it. The first USB devices were very slow compared to modern gaming devices. So I assume that the impact is less significant on lower polling rates, like when an office mouse is used instead of a gaming mouse.
This. I still keep my mouse set to 125 Hz on my mac laptop. At 1000 Hz, it may not be noticeable visually, but in terms of CPU usage and battery life it makes a sensible difference. Also something to note is that the HID driver (HID = USB keyboards and mice) is still using a lot of CPU cycles for each "event", still significant on a M1/M2 chip. On top of that, if you have any extension such as Karabiner, BetterTouchTool, etc. (or on Windows any soft setting a system hook on mouse events, or for messages in general) the OS will have to execute something in that program every single mouse event (meaning 1000 times a second instead of 125)! And it makes a noticeable difference in CPU usage and battery life. That's also why I had to choose a mouse brand which behaves properly at 125 Hz. My mouse has the "ENDURANCE" mode, so it's meant to be used at 125 Hz too. Many many other gaming mice I tried weren't designed for that, so they only behaved well at 500 Hz or more, which was a problem with laptops of around 2017.
@@florianb3935 Yes, as far as I know the OS typically converts the RAW input data into events which are typically handled in sequence by the application, likely causing a huge overhead. A mouse also likely waits some time before sending real input data to reduce the error. So unless a very high DPI is used the device didn't really send useful updates each (sub) frame. On my Razer Viper V2 Pro I some polls are skipped and often the movement data is only 1 unit and only after some polls that it sends a movement of 100 units because I only use 2300 DPI with 1 kHz polling.
Very interesting! I seem to remember not even encountering USB until 1996-1997 but don't remember really having any device to actually connect to it until 98-99 (hard to remember, it was a mouse though). It feels like the ports appeared before there was really much purpose for them which makes sense, it was just quite a slow (relatively) rollout since we didn't yet have cheap flash memory for a while which was the real killer app for USB. I don't recall any discussion of a performance hit back then but this makes sense since most of those ports went unused for a number of years, PS2 ports refused to die but it is genuinely surprising that these issues existed in the Pentium 2/AMD K6 era since this was when it was increasingly likely you would buy a USB keyboard/mouse. Do you have any newer platforms? would be curious to see if this was noticeable in 1999+ hardware like Pentium 3/Athlon platforms.
I'll definitely make a follow-up video using other platforms and CPUs. The PII I 1000 in this video didn't experience any performance hit, but you can see less performance in cache and memory throughput. It might not be enough to measure a difference though. I am currently fixing a P4 motherboard which I would also like to test.
This is old news. Though thank you for explaining it on other devices. I was told years ago the ps/2 keyboard interface was speedier than usb. As for reasons of ps/2 keyboard / mouse or combo port for so long esp. on business class machines is you didn't want to change everything out just because you changed out your pc. Esp large rows of KVM units. Thus, you would see all the legacy ports on a business workstation for years.
On my hardware I noticed that some motherboards are affected more than others with the same CPU, maybe different controllers and different BIOS can change that. What about not only using USB keyboard and mouse, but also USB storage and USB sound card all at the same time?
Good question - until now, I have only experienced the issue with input devices - in particular keyboards. USB thumb drives do not degrade performance. I am already collecting USB 2.0 PCI cards for another video. That would be the time to test different motherboards with the same CPU - and different USB controllers.
Oh thank goodness, the only USB input device i use is a Joystick... and the ill-fated Sidewinder Strategic Commander. I make it a point of pride to not use any USB keyboards or mice, only PS2 as Bill intended. With a video title like this, i was worried that the USB sticks i use for data transfer from my main PC to the retro PC would cause issues.
I would never use USB Input Devices on such an old hardware for standard usage. The performance is one side. Konventional Memory Usage another... The Emulation the BIOS provides to have USB Devices be used in DOS, is the worst.
I would expect this problem is bios specific. Some bioses will be better than others as it is going to be related to the Bios support for USB emulation of legacy input devices. The bios probably sets up a timer interrupt so it can poll the USB devices and that causes the performance loss. I would expect things might be different in Windows if Bios support for USB input is disabled as you are completely reliant on Windows USB HID support and I would expect differences between 9x and nt operating systems but with your retro pc you probably aren't using 2k or XP so nt performance doesn't really matter.
In Windows 98, I did notice a similar performance drop as in DOS. There was a difference though. In DOS, I could pretty much unplug the device and the issue immediately went away. In Windows, it took about 5 seconds. I'd assume it's related to the driver noticing that the device is no longer present, unloads and does some other cleanup.
Wow, the AMD K6-2 3D 300mghz was one of my... super fast chip Before that from the 6502 1mghz to the 8 then 30mghz 68030. jump up in speed. The fastest jump before that was perhaps my 68040 25mghz... and then 486 120mgz... AMD 3800 X2 cpus... The latest speed jump was from an AMD FX8350 to a Ryzen 7950x...
I wouldn't expect a converted USB mouse to PS/2 to have these issues. It's related to the port. As someone else pointed out, not all USB devices can be converted to PS/2. They need to support the PS/2 protocol, otherwise they won't work
i personally prefer using dedicated connectors when possible, i never trusted usb with its unreliable wiggly connector and its general purposeness. i much prefer using a dedicated driver and connectors with screws like db15 or some other locking mechanism like the little clips on hdmi. a few days ago my keyboard disconnected while playing a game. this made the system crash so hard i had to clear the bios settings and unplug the pc for a few minutes so it would turn back on. maybe it wasn't just the keyboard, but unplugging and plugging the keyboard back fixed the problem. the usb connector on my keyboard always fits lose on all connectors but as long as i don't move it this isn't a problem.
i would be somewhat interested is that indeed the cpu problem or a place where usb is located in the system , for instance p3 motherboard have integrated usb 1.1 but is it integrated a a pci device controlled by cpu or is it somewhat more advanced controller that delegates some features from cpu and for low level interupt devices like keyboard or mice that can be enabled as ps2 emulated mode in bios it might just skip asking cpu for usb instructions and just periodically reports to the system what it does with usb while routing interupt level devices trough system board controller instead so they functionally works as ps2/serial devices for compatybility with dos for instance so good test would be to use separate pci controller on all boards and using the same one and disabling usb hub integrated in the motherboard all together the other issue might come from how southbridge and north bridge communicate so testing via chipset talking with pentium 3 over pci bus vs intel qgr bus might change latency
In my new Socket 7 system (k6-2@400) I tried using Daemon Tools on 98SE to read game images through a USB 2.0 card. It made all my gameplay stutter whenever the disc was accessed. A physical CD didn't have any stutters even though my drive struggled to read it. Daemon Tools from the local IDE drive also played perfectly smooth.
@@bitsundbolts It's not really an issue, just another anecdote that USB can cause slowdowns in old PCs. I don't think it's hardware related but rather software related as I tried XP (or possibly 2k) on the same hardware and IIRC that worked smoothly. I would've kept using XP but 3dfx didn't release D3D drivers for the Voodoo 1 on this platform. I believe the particular card I was using has a VIA chipset. I think I have a card with an NEC chipset somewhere I could also try if I want a change in the setup.
I wish PCs still used a wide variety of expansion cards, nowadays it's just graphics card and WiFi card, and maybe a capture card or sound card once in a while if someone streams on console or is more of a sound nerd, but most builds will just have the first two, everything now being USB feels monotone and lifeless
When I tried using a USB keyboard with my K6-2, whenever I would press a key in Duke3D it would just grind to a crawl, then speed back up when I stopped. It was especially annoying because the keyboard PS/2 port on the motherboard was dead.
Also an interesting test. I have not tested this yet what would happen when I press keys during the cache benchmark - specifically with a PS/2 keyboard connected. I am wondering if interrupts would cause the same behavior as USB devices.
Connecting a single flash drive to my K6-2+ machine via USB2 card KILLED performance, way more than 3%! 3DMark2K score went from 1600 to 1200! The reduction is much less if I swap the USB2 card for USB1.1 but it's still there - maybe 10% or so in the same test. Also I did try USB2 cards from different manufacturers and there was some variance. IIRC the card with an nVidia chip did the best which is the one that dropped the score to 1200.
That is interesting since I couldn't notice a performance loss with storage devices connected to my systems. The behavior only appeared when using input devices like keyboards. It may come down to the USB implementation. I'll try to get different USB 2.0 expansion cards and see if I can recreate your experience.
@@bitsundbolts I believe it's something about that specific K6-2+ machine because I don't see that massive deficit with any other retro PC. I do recall seeing a SMALL deficit though, with my 440BX-based 1GHz Pentium III system when leaving a flash drive plugged-in.
There was an issue back in the day with the mouse moving causing the pc to actually lag up. People actually switched over to ps2 keyboards because of it.
IT IS noticiable in my k6-2 500 and Soyo 5EHM 1.3, when im using the mouse (Microsoft Intellieye 1.0 optical from 1999) you notice how the mouse start to stutter when you open things and the whole computer becomes much slower if you move data from a usb drive. Of course i disabled the usb 1.1 in the bios and i connected the mouse via PS2 as well the IBM Model M keyboard via PS2 to DIM adapter. Its the same thing with a VIA 2.0 PCI card
Just stay out of USB on motherboards where Super IO did not support it. That's anything with a slotted CPU and before. If you need flash storage, there are CF and microSD to IDE adaptors (or adapters, for US friends). However, these do require reboot to detect media changes. I have a K6 ][ with an SSD, via Vantec SATA to IDE card. You can buy them cheap, so it's good to have one around. Be sure you connect 3.5 floppy power plug for 5 volts. Have fun!😁
I was testing this with a NEC D720101GJ chip-based USB 2.0PCI card on an AMD K6III+ 550MHz @ 600MHz on an Asus P5A and noticed a 3% drop in composite system performance drop in Windows 98SE.
I haven't watched all the video yet, but if you don't know, not all USB chips are created equal. Have you compared OHCI with UHCI? OHCI USB implement a more intelligent chip while UHCI pushes more work to the CPU. Common vendors are... NEC for OHCI and Intel/ VIA for UHCI. I always avoid VIA USB cards. This applies to the USB 1.1 era and USB 2.0 cards running in 1.1 mode, but I'm not familiar with later stuff.
I was wondering, if you use a USB to PS/2 or 5-pin DIN converter, does the performance drop still happen or do they operate the same as PS/2 or does the polling problem still come up? I'd be curious to see if they still interrupt as often with the converters installed or if they interrupt like an older device.
When you use those adapters the device is just routing PS/2 signals though the USB cable as if it was a PS/2 mouse. There is no USB signals involved. Unless it is a fancy active adapter, which most aren't and even then the computer wouldn't see the USB protocol.
I'd argue a 2-3% performance loss is entirely negligible compared to the great convenience USB offers? I imagine if someone cared that a 60 seconds long pure CPU workload takes 62 seconds instead they wouldn't be doing retro stuff in the first place lol
eh I mean in certain end use cases USB can still be handy for transferring files if say for example you have a mechanical hard drive. I've used a thumb drive plenty of times with my DOS 6.22 Socket 7 system , it saves me giving up an expansion slot for a network card or having to remove my HDDs every time I need to put stuff on them. I do use PS/2 devices for my keyboard and mouse though.
USB is a convenient way to transfer files, I agree. There is one benefit I like using wireless USB keyboards and mice: less cables Since I change my setups a lot, it does help me keep my desk clean. But for benchmarks or long term use, it's probably better to stick with PS/2.
USB is very convenient for the end user but not so much for the developers. It requires entire protocol stack similar to that of bluetooth or networking. In "legacy" mode, simplified USB stack and legacy device emulation is implemented in the BIOS and runs in SMM (System Management Mode). Long story short, it's quite a lot of code that steals CPU cycles from the OS. Switching to and back from SMM takes time. And since it is using CPU caches, just like any other code, it causes even more performance loss. Not to mention that it is pretty much impossible to predict when SMM is going to kick in. Depending on how good or bad the implementation is, it might be pretty much invisible to the OS and the user or break everything entirely, including completely crashing the system. Updating the BIOS or changing the mainboard might change things by a lot.
I have seen SMM mentioned many times now in the comments. I may put the focus on it in the next video as "polling" may not have been the correct terminology.
Lately I've been trying out Settlers III for the first time as I've never had an opportunity to play the full game before, only a demo version back in the late 90s. Imagine my bafflement when I found out that the game freezes completely whenever I move the mouse around. As soon as I leave it still everything is fine again. How can I even play a real time strategy that way?? Surprisingly no such issue occurs when I use the trackpoint (the computer is an IBM ThinkPad T42 laptop with a Pentium M 1700 MHz; not all that retro, merely 18 years old). I have no idea what produces such weird symptoms. I wanted to give a PS/2 mouse a try and see if it works fine with the game but after closer inspection what I've hoped to be a PS/2 port turned out to be an S-video out so no cigar ;)
I've been capturing footage of Settlers III - for a new series about copy protections back then. I played it on a Pentium 200 with 96 MB memory. I even used a USB keyboard and mouse. It wasn't running well, but good enough to make it to Romans scenario 7. I have not encountered any such issues you describe. However, I once experienced a weird behavior in Jedi Academy (I think it was that game). I could play the game until a certain point, then, the frame rate went to 1 frame every 5 seconds. This was on a newer Intel Celeron N3450. I tried different video drivers and all kinds of other remedies. But I don't think it was related to input
I remember USB being so bad in the 90's that when I saw computers in 2006 that only had USB I'm like WTF!?!?!? Then I realized it wasn't bad anymore. Now I'm like.....why does my 2021 motherboard have PS2? lol
@BuB ich hätte da evtl. eine alternative Lösung für das USB-Problem. Ich habe Anfang diesen Jahres auch damit angefangen mich wieder mit alter Hardware zu beschäftigen. Mein "erster" Classic PC war ein HP Vectra 486. Leider stand ich schnell vor dem Problem, dass die Tastatur nicht komplett funktionierte (ein paar wenige tasten reagierten nicht) und ich dachte mir - das muss doch auch mit einer aktuellen Tastatur lösbar sein. Ich habe also einen kleinen USB auf ps2 Adapter entworfen. Funktionen aktuell: USB Maus oder Tastatur wird automatisch erkannt und auf PS2 Protokoll umgesetzt. Per Schalter lässt sich zwischen XT und AT Mode wählen (XT aber lediglich Tastatur Support). Falls du Interesse hast, könnte ich dir einen Adapter zukommen lassen sobald ich meine letzten Optimierungen mit eingebaut habe - vielleicht schaffe ich es auch noch bis dahin ein Gehäuse zu entwerfen ;). Ansonsten tolle Videos! Schaue mir gern deinen Content an und dafür, dass du in der Elektrotechnik lt. deiner Aussage "ein Laie" bist, sind deine Ergebnisse recht gut finde ich (bin gelernter Elektroniker) :).
Hallo! Das hört sich ja nach einem super Projekt an. Ich bin sehr an solchen Lösungen interessiert und würde gern ein solches Projekt/Produkt testen. Weiteres können wir gerne per E-Mail besprechen (bitsundbolts at Gmail Punkt com). Würde mich freuen!
1-3% seems lower than I would have expected. You wouldn't think that would be visible, but I remember it being visible to the naked eye (frame rates dropping).
I may have to run different benchmarks to get a better understanding of how bad the performance loss could be - especially games under windows. I'll also try to look at different USB chipsets and USB versions next.
You make a good point, and it is a great video, but if you are afraid of 1-3% speed losses, then maybe playing on a 20 year old computer is not for you? Whatever era you are trying to "cosplay", you will always be able to get top-of-the-line components that were faster than what you had access to in that era at the same price or cheaper (in respect to whatever the era in time is that invokes your melancholy response)
early implementationsv of usb was terrible i managed to get the usb breakout for an asus super socket 7 mobo with a k6-2 475 and the beta bios and the hacked together driver. its caused so much instability it only ever semi worked on xp sp2 lol
....my main question here - Who is tasked with the actual polling? Is that task mandated to the main CPU or does the USB controller contain circuitry for it...? I think, from the problem itself, that the CPU is included thus has a 'periodic leave' from its normal operation, hence the periodic dips in those graphs. But what happened with P3-1000? Is it just fast enough to make those interrupts invisible? Did it run in a different motherboard than the P2? What does that board show with a slower CPU, like a Mendocino Celeron? Maybe its USB controller is more advanced and took over from the CPU at last....? So, not one question after all, I'm full of questions.
Yeah, you just created content for another video right there. The PII was in a Gigabyte slot 1 motherboard. The PIII was in the Asus P3B-F using a slotket adapter. It could be that the difference in USB controllers makes a difference. I do have other slot 1 CPUs to dig even deeper. Maybe a good test while keeping the motherboard the same
The BIOS includes a emulation. On USB1 odr USB 2, the CPU has to handle the polling. Additionally it will take you some konventinal Memory, in most cases. I saw something beetweeen 8 or 16KiB used. USB Input devices should be avoided in general.
@@huberthans4312 8-16k? That's an inconveniently big chunk. I'll look into it (too) when I have some more modern piece on my test bench. One more reason to leave USB alone...
In this era, i have microlag when i review the stage in Colin McRae Rally 2001. I optimizing windows, not work, i rearranged the device IRQ-s, with exchange cards in the slots, and not work, but a little better. I have shared IRQ on the USB controller, and not able to change, but i only use USB when printing. Then disable USB in the BIOS, now the fps better 45 vs 54. This time i use Teac 8x CDRW, but this only PIO4 multiword DMA mode. It work super, and not cause any problem. I bought a faster Phillips 16x CDRW, and this UDMA66 capable, now the microlag not appear in the game. I later found a solution, a windows setting autoinsertcdnotification to disable and it work, no microlag with the old CDRW connected, but the new UDMA device never appear the microlag. I making test, when i set all device IRQ-s to 9, this cause -10fps in the game.
@@bitsundbolts I testing computers in this era, with Quake2 crusher.dm2 demo in loop for power and stability. I upgraded my cpu from Thunderbird 850 to Athlon XP1900. The fps jump from 30-32 to 38-40 in Quake 2 demo. I not interested. My little broder using Impulse Tracker, and we have AWE64 ISA card, and sound good, and very good compatibility, for every program. A little later i have little money, and bought a SB Live!, just for fun. It is compatible also. When updgrading the machine i select motherboard with ISA slot, to using AWE64, this time is not planned to change the sound card. Later coming the 3DMark with 3D sound test, and wow, this is good. I think, test the Quake with A3D setting, and mmm. Now test the Crusher.dm2 demo, with normal stereo, and with this sound card the fps jump from 38-40 to 55-58. Now i think, the modern computer using ISA device is a big brake, and if not enogh knoledge, the PCI also braking a little bit.
Interesting thought. I just wonder if it will be visible in the cache benchmarks. Might be something interesting to look at in another follow-up! Thanks!
Yes, I have tested this, but I did not notice any difference. Having no USB input device connected is equal to disabling USB in the BIOS. At least this was the behavior of the hardware and benchmarks I used.
ofc USB didn't affect pentium 3. Pentium 3 was a big jump in performance in Coopermine and even more in Tualatin cores. Pentium 3 are very fast machines, i love them.
Yes, I have noticed this large drop as well. As you said, probably not enough to notice a difference in the benchmarks I used. Maybe there are others where it is noticeable again.
i know retro pc's are a thing but ....this is like saying you can over clock a raspberry pi 5 . ANY of these issues can be resolved by UPGRADING to a faster computer with more modern i/o
I have some old HDD arriving soon. If they work, I will make some HDD videos in the coming months as well. And a great idea - my retro systems are snappy and fast - and that is probably because of the ide-to-sd card adapter. I already feel the pain getting into Windows when using one of those old HDD!
I have several retro computers that have usb and none of them have any issues with pendrives or keyboard and mouse. I just use usb in order to use pendrives but I ran tests with a wireless keyboard and mouse and everything was ok. No change in performance.
That's interesting. 3 of the 4 boards I have tested in this video show the issue visually in SpeedSys when an input device like a keyboard is connected via USB. Could you share the hardware you're using?
Nice video. But you still don't include the USB controller type to comparison. Via chiset is using uhci controller which is the worst possible one. Ohci controllers are much better and will result in less cpu usage.
I'm in the process to source different PCI USB expansion cards (USB 1.1 and USB 2.0). I hope to be able to reproduce the experience many commenters share.
I have to look into this more in depth. I don't think that the two socket 7 motherboards have that option in the BIOS. They only allow to enable or disable USB. I may test this with PIII and maybe even a P4 system
At the moment, I would say yes. I tested wired and wireless HID devices, and they all show this behavior. The issue is not present with USB storage devices like a thumb drive.
So only usb 1 is an issue because it doesn’t have its own hardware to use. So usb 3 should be used cause it has its own chips to process data independently of the system itself?
Probably not. If the USB device is capable of communicating via an adapter to the PS/2 bus, then it implements the correct protocols. In that case, it would behave like an ordinary PS/2 device
I did disable the USB controller in one test which I did not include in the video. It is identical to no input device connected to the ports. So, disabling the USB feature in the BIOS does not impact the system.
The start wasn’t exactly great, made it sound USB came and replaced FireWire, were two different types of connections back in the days, FireWire for high speed data transfer and USB for slower like printers mouse keyboards and FireWire for video external storage Overtime they increased the USB data transfer speed and eventually FireWire became less used due to certain factors like majority of PCs just had USB unlike Mac PCs and license
Yes, you're right. But there was a problem with FireWire - not many PCs had it built in. I remember that I required an expansion card to have FireWire available. Later, a FireWire port was more common to be on the motherboard. You are correct that FireWire was the way to go for storage devices until later USB revisions.
Before usb you had Atari sio, literally usb before usb, native plug and play, supplies power, serial data, hardware handshakes, supports hubs and daisy chaining, and hardware controlled transfer speed up to 128kbyte/sec So you can write all 512k ram in atari computer (if u have ram upgrade past xe and 64k) in about 5 seconds
Fascinating! I had no idea about this issue.
Hello Adrian! Thanks for passing by and watching my video. I am glad you found it interesting!
The problem with the older USB Standards is that everything HAD to pass through the CPU during USB transfers. Something that thankfully was finally abandoned with the introduction of USB 3.0 and 3.1. Which i think is definitely the biggest gain from this standard upgrade, beside the file transfer speed.
This is also the reason why Firewire was superior even to USB2, even when it looked sighly slower at specs. Firewire controllers, just as SCSI before them, were always "smarter" than their USB counterparts. The CPU in the FW ASIC of the card and the peripherals could do more by itself and transfer the data between them and the host by using DMA with almost no CPU intervention. Also, the software stack was smaller, which meant you could do things like implementing Boot Support, Debugging Mode and Target Mode (your computer appears as an storage device to another computer) in firmware in a very compact way.
Ofc, since firewire controllers were almost microcomputers by themselves, that means the controller chips for both cards and peripherals were more complex, expensive and less scalable to produce than USB counterparts, which in the end was firewire demise.
Just want to say, USB overhead was well known at the time and even latency was an issue. This was all known and tested extensively by nerds a long time ago. USB didn't really become a non-issue until USB 2.0. I remember very well discussing this ad nauseam on forums back in the day.
It looks like USB 1.0 generates interrupt for CPU every time mouse moves (instead using DMA bypassing CPU entirely).
Every CPU handle interrupts and context switching with different penalty in cycles lost. Modern CPU has special HW to accelerate context switching due to those were designed for modern multitasking OS. Obviously old CPU designed for single tasking DOS lacks this support.
It will be really interesting to check how different usb 2 PCI controllers affect performance. Nice video!
Thank you! Will try to get some PCI USB 2.0 cards with different chipsets. I am curious too!
@@bitsundbolts I just bought that identical ALi USB2.0 Card! It is a nice performer on my i440bx board.
@@bitsundboltsThis was going to be my suggestion also. Various USB chipsets may show different results. On some more expensive USB ISA/PCI cards each port might get its own controller. Also drivers in various OS's could be a factor.
Also, did you see any difference with a wired USB device and wireless? 🙂
I did not test with a wired USB keyboard this time, but in my first video, there was no difference between a wired and a wireless keyboard. I would assume they behave the same.
@@bitsundbolts That's interesting and good to know. I suspected that the wireless receiver might be more active and sending a lot more USB traffic 🙂👍
People might be wondering why soundcards had all those extra ports and not the video cards or something else. Well, in the early days, soundcards were primarily used for gaming,. There wasn't much else that would use sound card, we didn't have music like we do today. It was mostly noise, triangle waves, and such. Multimedia didn't really come to computers until the early 90s. Computers like the Amiga were popular for making music, but it was still quite challenging to store CD-quality audio tracks on your system. Back then, we had about 40-80MB of HDD space, if I remember correctly. Companies knew that if someone bought a soundcard, it was likely to play games.
Great video again! 👍
Thank you! And interesting information which makes a lot of sense! Funny enough, I remember most of the soundcards I had back when onboard audio wasn't really a thing. The first soundcard ingot was an ESS688 AudioDrive - it was such an essential piece of Hardware in my AMD 486-DX4-100.
Aside from puny HDD sizes, Amiga's sound chip (Paula) from 1984 was never updated and the it was stuck with its 8-bit DACs till the end. That made it unfit for CD-level quality audio. Generally, the platform was stagnant. Most people bought cheaper 'small Amigas' that lacked Zorro slots to put expansion cards and have more PC-like market for graphic cards (the so-called ReTargettabble Grpahics or RTC - ability to use graphic cards appeared very late, only for 'big' Amigas with expansion slots and - for obvious - reasons only for well-behaved Workbench GUI apps, not for games). To make matters worse, games bypassed the OS and talked to hardware directly for performance, so most targeted the most popular config: a sub 1 MB Amiga 500. That made Amiga a hostage of her (rapidly aging) original chipset: updates were minor to not break compat and due to Commodore skimping on R&D.
13:14 I think I had this exact A4tech serial mouse. Thanks for bringing back the memories!
As for the USB performance drop, try external card with a NEC chip. I think I read somewhere that it has lower overhead in comparison to USB cards on a VIA chip.
I'm now installing a serial port on my Ryzen 7 5700X. Might be a tad slower for transferring videos from the iPhone. Next video is then scheduled for 2027! Thanks for the comprehensive analysis!
Retro PCs Tony, Retro PCs :) Please do not postpone your content until 2027! It would be a shame to deprive your community from your great content for so long!
I don't know if it's just my modern perspective of CPUs and USB protocol, but I don't quite understand how a slight performance dip of 1-2% on an older system can make a difference in the long run. But this is excellent information to help me learn how it can have an effect, and I'm glad you went through the effort of testing this. Your videos have been amazing to watch through, from your 3DFX GPU mods to the AMD K6-2+ mod that got me started with your channel. Please, keep working hard on this content, because I hope to learn more in the future.
Thank you for your kind words 🙏
Something to consider is that since a lot of the games at the time were targeting sub-30 FPS and that most games had their speed directly corresponding to how fast your CPU was (meaning that a faster CPU literally translated into faster game logic, thus the emulation problem were games were running "too fast"). 1-2% difference in performance is certainly meaningful.
For a modern comparison, a 2% difference in performance from a game running at 60 FPS could literally translate into a loss of 1.2 FPS which can wreak havoc if you're trying to achieve a 60Hz VSync frame lock.
Imaging trying to speedrun a game and then finding out that the mouse and keyboard you have connected changes how fast your car can literally go, or how fast your character can literally run, etc.
Sure it's not "insane" performance drops but it's certainly enough to look into (especially considering that we may see even more issues when using expansion cards or moving to USB 2.0, but hopefully BaB can give us that video in the near future to see for ourselves)
I think a much more real world problem is CPU load during USB bulk transfer. This affects any application that requires CPU speed and fast data transfer at the same time.
When editing video on a Mac in ~2005 on an external HDD it was strongly recommended to use FireWire instead of USB because FireWire used DMA and caused much less CPU load.
The difference could be dramatic.
Of course, a few years later faster processors have made that problem mostly irrelevant.
Maybe not irrelevant - because developers have used the extra processing power to implement more complex real time effects in various editing suites, while the need for latency-free recording is as important as ever.
Firewire was nice but e-SATA was better. RIP e-SATA though, when USB 3.1 came out it pretty much made e-SATA pointless. I wish we could have kept e-SATA honestly.
I've added several USB 2.0 cards to my retro rigs but only for quick file transfer. Good to see that I made the right choice keeping my input devices analog.
Interesting thing about adding usb cards is the PCI version. You can't just add any generic card because chances are that it uses a higher PCI version than is supported on your old motherboard. There's a pretty good thread over on Vogons about which chipsets to pick and avoid.
Great! Thanks for the hint. I'll have a look before I get useless PCI cards!
PS/2 and parallel ports were by no means analog, only sound and VGA ports used analog signals.
there is a mod to connect 3.3 volts make modern usb cards to work.
The performance drop is waaay bigger than 4%, I remember back in the day I had a PIII 550 Mhz, and when I switched from a PCI 56K to an USB 512 Kbps modem, when playing Counter-Strike, my FPS went down from 50+ to low 30's, it was immediately noticeable, I had to keep it unplugged all the time to just play single player games, I begged my parents for months to switch to another ISP, when they finally got an RJ-45 ethernet gateway it was such a relief, I got back on good form again.
Even today in modern 8 core systems, you can clearly see in task manager that when moving a 1000 Hz "gaming mouse" prevents the CPU from getting into the lowest power and clock states, my i7 4770 can downclock to 1 Ghz when moving a PS/2 at 200 "Hz" (reports per second) or 125 Mhz USB mouse, but when having a 1000 Hz + USB microphone, It idles at around 1.3-1.5 Ghz... that is the reason even today, laptop touchpads and keyboards still use the PS/2 protocol, to save CPU power.
It would be cool to re-run the tests with the USB devices actively in use and not just idling, like using the USB soundcard from your headset + KB and mouse moving, there are other things that also affect CPU performance on retro (and modern...) PC's, using the Realtek onboard soundcard (specially if you enable the background noise suppression), and high amounts of open TCP/IP connections, write caching on HDD's, having Hyper Threading enabled when not taking advantage of the extra threads, and some more I don't remember....
Very interesting video!
A few details and thoughts:
- From the PC's perspective it makes no difference if the USB supports "real" interrupts or not: The USB host controller manages the USB protocol side (including polling the devices) and sends a (real) interrupt to the PC whenever necessary. Some controllers might send unnecessary interrupts and let the driver figure that out, in which case you would have a significantly higher interrupt load. I'm not aware, however, how frequent that behavior is in practice, so let's keep that in the "theories" drawer. On the other hand the fact that on some platforms you tested there were periodic dips in the performance, presumably without any mouse movement, would support that hypothesis.
- The problem presumably comes from the fact that the USB driver stack is interrupting the current program and affects the contents of the cache - just as the PS/2 interrupt would - but is probably more complex, accesses more data and affects the cache more than legacy input device drivers.
- The specific driver implementation is another issue: If not in legacy emulation mode, the USB communication goes through the driver the manufacturer of the host controller provided. I suspect the implementations will differ wildly. It would be interesting to see how this issue affects Linux, which I suspect might have more consistent and better quality drivers overall.
- It would also be interesting to compare different host controller cards, and specifically USB 2.0 ones.
- As was mentioned in another comment, legacy emulation could lead to poor performance as well and needs to be tested and compared on a motherboard level basis to understand the impact.
The biggest factor making a difference is the USB controller being used. Some are more heavy in interrupting the CPU than others, who might to more things in hardware before interrupting software.
I am not surprised, this was one of the overclocker arguments from Kingpin when making the EVGA's Socket 1366 motherboards and offering IDE connector for the harddrive and PS/2 for mouse and keyboard.
Basically simplify and reduce parasitic loads for benchmarking runs and get that higher score.
Of course if you run a less powerful CPU, the performance costs start to become noticeable.
Interesting that it's up to Socket 1366 for this to be an issue. I guess if you're chasing world records, you want to have every little extra performance you can get!
USB penalty in retro devices isn't just hardware issues, but also software issues. The USB stack modules included in BIOSes for KB/Mouse support in DOS mode aren't smaller things and can consume precious UMB blocks you would need for other stuff. Sometimes it will prevent EMS memory from working as the USB BIOS module will steal the EMS window memory block (some Systemsoft BIOSes show the problem, it can be fixed assuming the board manufacturer provided a fixed update for that). This enables SMM/NMI which can conflict with DOS software and DOS drivers which also make use of NMI (sound card drivers or emulators for example).
In Windows, USB will always enable WDM (NTKern). Depending in your board and devices, it can conflict with VxD drivers, mostly sound cards. It will consume more memory, and make the OS slower in some circustances.
It just looks so damn cool when you have L3 cache with a K6-based platform.
Amazing video as always! I've appreciated same behaviour as you since you published the previous video showing this, when tested it on a 166MMX machine.
But at least for me, having the possibility of having my childhood pc now connected to a big flat tv and wireless keyboard&mouse clearly compensates the 1% penalty! Haha :)
Haha, yeah, it's not that big of a deal. But still weird that we can see a cache performance issue when a USB input device is connected.
Nice to hear that you tested this and the behavior is the same for you. Give a bit more credibility to my videos 😉
It was known 20 years ago that USB was/is CPU bound. One of the flame war arguments back then was that firewire was not...... USB was soo convenient that the 1 to 2% performance loss was acceptable to the majority of users. Nowadays it is hard to find even PS2 new keyboards/mice. I just bought a no-name PS2 keyboard from Amazon for an Amiga motherboard with DB9 (Amiga) and PS2 ports, but no USB, I also bought a little USB controller chip which sits on the CIA chips to provide USB keyboard support. AmiKey1200. Cool video. Never before had I seen the effect of USB being CPU bound.
This video was timed perfectly - I may end up needing to use my mouse in USB mode on a baby AT board that I recently purchased as it is missing the PS/2 connector from its LPT bracket, and the brackets from my other ("modern") baby AT board uses a different connector. The system will be running a Celeron 333 (or a Pentium II 233/350) under Windows 3.1 or 95 so quite relevant knowledge.
Glad to hear that the video may give you some insights. Maybe you can run SpeedSys to verify the behavior of the USB device and if it affects the performance.
@@bitsundbolts Considering the performance of a Celeron (with cache) is near on par with a same-clock Pentium II, and with the board (PCPartner VIB862D w/SB Vibra16XV) having the Via VIA VT82C691 and VIA VT82C586B, the results should be accurately comparable.
I could test with a Celeron 300A/450A (contact B21 cut) and PIII 650 if you want - both on my new board (since I can select the CPU speed, and I fully expect relatively low performance) and Intel SE440BX-2 (currently setup next to me).
Can't reliably test my current PII 350 as its heatsink is too large and it barely fits my boards, and the PII 233 only posts into BIOS (and only at 133MHz with the BIOS jumper set to "config").
I also have a Celeron 500 and 433 that could be tested in a slotket - they work flawlessly on the Intel board.
If you find the time, it would be interesting to see if you can replicate the behavior in any of your retro setups. I do have a few Celeron CPUs here as well, but I haven't tested those yet. I can imagine that the USB issue is present in almost all builds from that time up to around 700/800 MHz
I'll have a go, realistically I shouldn't need to do more than disable the memory manager to get real mode working for the tests. Might also give my mobile Duron 1300 a go on my GA-7IXEH, as it only runs at 500MHz on it and the board has the as far as I can tell unique to it Via KT133E.
I'm not going to use my Jetway994 as I still haven't gotten it reliable enough to comfortably say any result is accurate (I've had back-to-back CPU benchmarks vary by over 25% with a barebones setup and only PS/2 mouse and keyboard - and it is unstable as well).
@@bitsundbolts Starting the tests now with the PIII 650, commented out everything in autoexec+dosstart (also copied then cleared config.sys) and started the tests. I just need to remember to enable legacy USB support in the BIOS (I've had no need for it due to using an adapted MS Wheel Mouse Optical and a Fellowes adapted AT keyboard with my analogue KVM).
Nice findings! Please, next time pay the attention if USB Interrupt and/or USB Legacy Support are ON/OFF in the BIOS setup. This may also do a measurable impact on the results.
This! I was thinking watching this video if "Assign IRQ For USB" and "Legacy USB Support" options in the BIOS make a difference. I stopped using USB mice or keyboards on my 90s PCs years ago since it affected the sound output, I could clearly hear interference when using them, no matter what.
Turns out I had the USB vs PS2 inner workings wrong but with more or less the same result XD.
Someone told me that PS2 was better because it sends interrupts when it needs to, so you get "no latency" on games. , but USB sends a request and joins the scheduler of the cpu so your input can be "delayed".
Your explanation is clearly the correct one.
Probably someone noticed a difference when gaming but didn't made the proper tests to see a performance drop.
Interesting observations. I've been using wireless USB input devices on an old P233 MMX system as it is a lot more convenient than using an old crusty DIN keyboard and PS/2 mouse. At least the performance degradation isn't too severe, so the added convenience of USB might be worth the slight performance hit.
Yes, the performance hit is small and won't make a difference if some application or game is already running poorly. Convenience is a very good argument, especially when I don't have to have all those cables around!
It's interesting that you reported this. I have a VIA USB 2.0 PCI card in my PIII-933 system. I use it routinely for large file transfers as I have reliability issues with my optical drives. I also don't expose my current systems to my retro systems over the same network. Honestly, it doesn't affect performance for me to notice it under Windows 98 SE (USB storage extension) and 2000 Pro.
I have not seen any issue with data transfer (HDD or flash drives). The issue only appears when using input devices like keyboards and mouse. I will get some PCI USB 2.0 expansion cards with different chipsets (NEC, ALI, VIA, etc.) and see what is going on there.
I think the reason you are getting cache performance degradation is the bios regularly trapping into SMM for legacy usb support via a timer SMI, and SMM code running from a different location, accessing different data will surely trash your cache.
You may be right, unfortunately, I do not have the technical knowledge. All I can say is that this behavior was present in two socket 7 boards and one slot 1 board. The other slot 1 (P3B-F + PIII 1000) did not show such behavior which may have been due to the fast CPU, but I have to test this board with a slower CPU once more.
Does this mean if the bios has the option to disable legacy usb support it might not show a negative effect in windows?
It could be - I will try to include this test in the next video I make about this topic.
Using a PS/2 adapter would also bei interesting.
Thank you for investigating deeper in this topic!
I think you mean a USB-to-PS/2 connector. I would assume it to behave like a regular PS/2 device, but I have not tested this yet. It would turn a "bad" USB keyboard into a "good" PS/2 keyboard.
@@bitsundboltspassive adapters require the mouse or keyboard to support ps/2 protocols. I use microsoft wmo 1.1a with passive usb to ps/2 converter. At the bottom of the mouse “usb and ps/2 compatible” is written. Meaning it would work just like a native ps/2 device.
Oh yes, I remember being stuck because of this! You can't just turn any USB device into a PS/2 device. It must be compatible and support the protocol (when using passive adapters)!
Thanks for reminding me 😀
@@bitsundbolts yeah! It was always a hit or miss. I have had noname usb keyboards that dont mention ps/2 compatibility work with ps/2 converter.
@@bitsundbolts exactly! The requirement for being campatible with the protocoll is also a good hint. But still I think using an adapter could help in widening the available peripherals i.e. the standard but "good" cherry keyboards.
Great research! Would be interesting to get an overview of percentages performance loss across core frequencies, especially with the P2 and P3 era. Wondering if it would be linear with clock speed, and at what point it's no longer measureable.
With each video I make about this topic, I have more questions afterwards once I read the comments. Although I have a feeling that around 800 MHz, it will be difficult to measure a difference, it could also be that I have to change benchmarks and testing methodology. For now, it's great to see the graphs in SpeedSys of the cash - at least we know something is there.
I believe results can vary based on controller chip or other factors, but particularly with add in USB2 cards, that the USB controllers being enabled (no device plugged in) in device manager can at times be enough to cause a performance degradation. Of course that means you can just toggle the controller on and off as needed if max performance is desired. I have gotten mixed messages from others on the exact conditions on when the degradation happens, but I know what works for my system.
Plenty of fun stuff to test there since USB 1.1 comes in OHCI and UHCI flavors, and USB 2 cards expose one of those on top of the EHCI controller which can be disabled individually. If my memory serves they also as a result, if all controllers are enabled, use 3 of the 4 PCI IRQs, so while not a huge deal IRQ sharing could be a factor in achieving the last bit of performance.
I still use a PS/2 keyboard with my modern system. I would use a PS/2 mouse as well if my motherboard had a second PS/2 port. It only has one PS/2 for keyboard OR mouse. PS/2 ports have 6 pins but only 4 pins are used. On motherboards with only one PS/2 port, the 2 unused pins should be the clk and data for a second PS/2 port via a Y adaptor. All the super I/O chips have 2 PS/2 ports so there is no reason why the clk and data for the second port couldn't be routed to the 2 unused pins on motherboards with only one physical PS/2 connector.
Very interesting, would have loved to know this back then. Another video topic I would like to see is what would a late model 486 with PCI.
Good job, thanks.
Just a few days ago i was testing an old system equipped with a Pentium 3 667MHz on an AOpen AX34 Pro, 256 MB of SDR PC133 and a Voodoo 3 3000 running windows 98SE. That system has only 2 USB ports in the rear I/O panel so i added a VIA based pci usb card. With a usb stick connected to that card, running 3D mark 99 Max i was getting a final score that was half the result i previously managed to get...unplugged the usb stick and scores were back to normal... Pretty extreme behaviour, i know that VIA pci usb cards are doomed but this behaviour is fare beyond my expectations...
The older connectors are interrupt based, meaning the CPU only needs to do something if there was an input. USB needs to be polled regularly at the rate the USB device tells the host to be in the endpoint descriptors. That the polling rate can have a negative impact isn't just an old problem. Just four months ago Microsoft released an update for high polling rate issues on Win 11. As far as I know there were a lot of issues too back then when the first 1 kHz mouses came out. So not even up to date systems can really properly handle high polling rate devices, so no wonder that on older system can't deal with it.
The first USB devices were very slow compared to modern gaming devices. So I assume that the impact is less significant on lower polling rates, like when an office mouse is used instead of a gaming mouse.
This. I still keep my mouse set to 125 Hz on my mac laptop. At 1000 Hz, it may not be noticeable visually, but in terms of CPU usage and battery life it makes a sensible difference.
Also something to note is that the HID driver (HID = USB keyboards and mice) is still using a lot of CPU cycles for each "event", still significant on a M1/M2 chip. On top of that, if you have any extension such as Karabiner, BetterTouchTool, etc. (or on Windows any soft setting a system hook on mouse events, or for messages in general) the OS will have to execute something in that program every single mouse event (meaning 1000 times a second instead of 125)! And it makes a noticeable difference in CPU usage and battery life.
That's also why I had to choose a mouse brand which behaves properly at 125 Hz. My mouse has the "ENDURANCE" mode, so it's meant to be used at 125 Hz too. Many many other gaming mice I tried weren't designed for that, so they only behaved well at 500 Hz or more, which was a problem with laptops of around 2017.
@@florianb3935 Yes, as far as I know the OS typically converts the RAW input data into events which are typically handled in sequence by the application, likely causing a huge overhead. A mouse also likely waits some time before sending real input data to reduce the error. So unless a very high DPI is used the device didn't really send useful updates each (sub) frame. On my Razer Viper V2 Pro I some polls are skipped and often the movement data is only 1 unit and only after some polls that it sends a movement of 100 units because I only use 2300 DPI with 1 kHz polling.
Very interesting!
I seem to remember not even encountering USB until 1996-1997 but don't remember really having any device to actually connect to it until 98-99 (hard to remember, it was a mouse though). It feels like the ports appeared before there was really much purpose for them which makes sense, it was just quite a slow (relatively) rollout since we didn't yet have cheap flash memory for a while which was the real killer app for USB.
I don't recall any discussion of a performance hit back then but this makes sense since most of those ports went unused for a number of years, PS2 ports refused to die but it is genuinely surprising that these issues existed in the Pentium 2/AMD K6 era since this was when it was increasingly likely you would buy a USB keyboard/mouse.
Do you have any newer platforms? would be curious to see if this was noticeable in 1999+ hardware like Pentium 3/Athlon platforms.
I'll definitely make a follow-up video using other platforms and CPUs. The PII I 1000 in this video didn't experience any performance hit, but you can see less performance in cache and memory throughput. It might not be enough to measure a difference though. I am currently fixing a P4 motherboard which I would also like to test.
This is old news. Though thank you for explaining it on other devices. I was told years ago the ps/2 keyboard interface was speedier than usb. As for reasons of ps/2 keyboard / mouse or combo port for so long esp. on business class machines is you didn't want to change everything out just because you changed out your pc. Esp large rows of KVM units. Thus, you would see all the legacy ports on a business workstation for years.
i did enjoyed the content , as always .
I remember my USB mouse causing slight lags at 500Hz pooling on Pentium 4@3GHz, lowering to 250Hz solved the slowdowns.
On my hardware I noticed that some motherboards are affected more than others with the same CPU, maybe different controllers and different BIOS can change that. What about not only using USB keyboard and mouse, but also USB storage and USB sound card all at the same time?
Good question - until now, I have only experienced the issue with input devices - in particular keyboards. USB thumb drives do not degrade performance. I am already collecting USB 2.0 PCI cards for another video. That would be the time to test different motherboards with the same CPU - and different USB controllers.
Oh thank goodness, the only USB input device i use is a Joystick... and the ill-fated Sidewinder Strategic Commander. I make it a point of pride to not use any USB keyboards or mice, only PS2 as Bill intended.
With a video title like this, i was worried that the USB sticks i use for data transfer from my main PC to the retro PC would cause issues.
I would never use USB Input Devices on such an old hardware for standard usage. The performance is one side. Konventional Memory Usage another... The Emulation the BIOS provides to have USB Devices be used in DOS, is the worst.
Interesting - I did not even think of memory usage for USB emulation.
I would expect this problem is bios specific. Some bioses will be better than others as it is going to be related to the Bios support for USB emulation of legacy input devices. The bios probably sets up a timer interrupt so it can poll the USB devices and that causes the performance loss. I would expect things might be different in Windows if Bios support for USB input is disabled as you are completely reliant on Windows USB HID support and I would expect differences between 9x and nt operating systems but with your retro pc you probably aren't using 2k or XP so nt performance doesn't really matter.
In Windows 98, I did notice a similar performance drop as in DOS. There was a difference though. In DOS, I could pretty much unplug the device and the issue immediately went away.
In Windows, it took about 5 seconds. I'd assume it's related to the driver noticing that the device is no longer present, unloads and does some other cleanup.
I'm betting my Pentium II Overdrive (Pentium Pro Upgrade) will have the same dip as the slot 1. Noted. Thank you for the video
Wow this is very interesting!
Wow, the AMD K6-2 3D 300mghz was one of my... super fast chip
Before that from the 6502 1mghz to the 8 then 30mghz 68030. jump up in speed.
The fastest jump before that was perhaps my 68040 25mghz... and then 486 120mgz... AMD 3800 X2 cpus...
The latest speed jump was from an AMD FX8350 to a Ryzen 7950x...
Great video and had no idea. I'm wondering for a USB mouse using a PS2 converter if it would have no effect since it's still using PS2 connection?
I wouldn't expect a converted USB mouse to PS/2 to have these issues. It's related to the port. As someone else pointed out, not all USB devices can be converted to PS/2. They need to support the PS/2 protocol, otherwise they won't work
I used PS2 for a very long time, up to 10th gen Intel. It's a shame they removed it, even if it's just to make more USB ports available.
i personally prefer using dedicated connectors when possible, i never trusted usb with its unreliable wiggly connector and its general purposeness. i much prefer using a dedicated driver and connectors with screws like db15 or some other locking mechanism like the little clips on hdmi.
a few days ago my keyboard disconnected while playing a game. this made the system crash so hard i had to clear the bios settings and unplug the pc for a few minutes so it would turn back on. maybe it wasn't just the keyboard, but unplugging and plugging the keyboard back fixed the problem. the usb connector on my keyboard always fits lose on all connectors but as long as i don't move it this isn't a problem.
i would be somewhat interested is that indeed the cpu problem or a place where usb is located in the system , for instance p3 motherboard have integrated usb 1.1 but is it integrated a a pci device controlled by cpu or is it somewhat more advanced controller that delegates some features from cpu and for low level interupt devices like keyboard or mice that can be enabled as ps2 emulated mode in bios it might just skip asking cpu for usb instructions and just periodically reports to the system what it does with usb while routing interupt level devices trough system board controller instead so they functionally works as ps2/serial devices for compatybility with dos for instance
so good test would be to use separate pci controller on all boards and using the same one and disabling usb hub integrated in the motherboard all together
the other issue might come from how southbridge and north bridge communicate so testing via chipset talking with pentium 3 over pci bus vs intel qgr bus might change latency
In my new Socket 7 system (k6-2@400) I tried using Daemon Tools on 98SE to read game images through a USB 2.0 card. It made all my gameplay stutter whenever the disc was accessed. A physical CD didn't have any stutters even though my drive struggled to read it. Daemon Tools from the local IDE drive also played perfectly smooth.
It could be related to the USB chipset used on the card. I will look into this, maybe I'll be able to reproduce your issue.
@@bitsundbolts It's not really an issue, just another anecdote that USB can cause slowdowns in old PCs. I don't think it's hardware related but rather software related as I tried XP (or possibly 2k) on the same hardware and IIRC that worked smoothly. I would've kept using XP but 3dfx didn't release D3D drivers for the Voodoo 1 on this platform.
I believe the particular card I was using has a VIA chipset. I think I have a card with an NEC chipset somewhere I could also try if I want a change in the setup.
I actually used a parallel port a couple days ago to dump a rom.
Wish modern computers still had them!
I wish PCs still used a wide variety of expansion cards, nowadays it's just graphics card and WiFi card, and maybe a capture card or sound card once in a while if someone streams on console or is more of a sound nerd, but most builds will just have the first two, everything now being USB feels monotone and lifeless
When I tried using a USB keyboard with my K6-2, whenever I would press a key in Duke3D it would just grind to a crawl, then speed back up when I stopped. It was especially annoying because the keyboard PS/2 port on the motherboard was dead.
Also an interesting test. I have not tested this yet what would happen when I press keys during the cache benchmark - specifically with a PS/2 keyboard connected. I am wondering if interrupts would cause the same behavior as USB devices.
Nice video!
It also matters what the polling rate of the USB device is. Modern peripherals are reaching as high as 8 Khz, which would probably melt a Pentium II.
I'm glad that at least the presence of the USB port is not the issue, so we can still use USB for file transfers.
Yes, absolutely!
Connecting a single flash drive to my K6-2+ machine via USB2 card KILLED performance, way more than 3%! 3DMark2K score went from 1600 to 1200! The reduction is much less if I swap the USB2 card for USB1.1 but it's still there - maybe 10% or so in the same test. Also I did try USB2 cards from different manufacturers and there was some variance. IIRC the card with an nVidia chip did the best which is the one that dropped the score to 1200.
That is interesting since I couldn't notice a performance loss with storage devices connected to my systems. The behavior only appeared when using input devices like keyboards. It may come down to the USB implementation. I'll try to get different USB 2.0 expansion cards and see if I can recreate your experience.
@@bitsundbolts I believe it's something about that specific K6-2+ machine because I don't see that massive deficit with any other retro PC. I do recall seeing a SMALL deficit though, with my 440BX-based 1GHz Pentium III system when leaving a flash drive plugged-in.
There was an issue back in the day with the mouse moving causing the pc to actually lag up. People actually switched over to ps2 keyboards because of it.
IT IS noticiable in my k6-2 500 and Soyo 5EHM 1.3, when im using the mouse (Microsoft Intellieye 1.0 optical from 1999) you notice how the mouse start to stutter when you open things and the whole computer becomes much slower if you move data from a usb drive. Of course i disabled the usb 1.1 in the bios and i connected the mouse via PS2 as well the IBM Model M keyboard via PS2 to DIM adapter. Its the same thing with a VIA 2.0 PCI card
gotta love that MV700i
Just stay out of USB on motherboards where Super IO did not support it. That's anything with a slotted CPU and before. If you need flash storage, there are CF and microSD to IDE adaptors (or adapters, for US friends). However, these do require reboot to detect media changes. I have a K6 ][ with an SSD, via Vantec SATA to IDE card. You can buy them cheap, so it's good to have one around. Be sure you connect 3.5 floppy power plug for 5 volts. Have fun!😁
I was testing this with a NEC D720101GJ chip-based USB 2.0PCI card on an AMD K6III+ 550MHz @ 600MHz on an Asus P5A and noticed a 3% drop in composite system performance drop in Windows 98SE.
With the time period we are talking about. I'm not surprised at all. Old computers are slow.
I haven't watched all the video yet, but if you don't know, not all USB chips are created equal. Have you compared OHCI with UHCI? OHCI USB implement a more intelligent chip while UHCI pushes more work to the CPU. Common vendors are... NEC for OHCI and Intel/ VIA for UHCI. I always avoid VIA USB cards. This applies to the USB 1.1 era and USB 2.0 cards running in 1.1 mode, but I'm not familiar with later stuff.
I was wondering, if you use a USB to PS/2 or 5-pin DIN converter, does the performance drop still happen or do they operate the same as PS/2 or does the polling problem still come up? I'd be curious to see if they still interrupt as often with the converters installed or if they interrupt like an older device.
When you use those adapters the device is just routing PS/2 signals though the USB cable as if it was a PS/2 mouse. There is no USB signals involved. Unless it is a fancy active adapter, which most aren't and even then the computer wouldn't see the USB protocol.
I'd argue a 2-3% performance loss is entirely negligible compared to the great convenience USB offers? I imagine if someone cared that a 60 seconds long pure CPU workload takes 62 seconds instead they wouldn't be doing retro stuff in the first place lol
eh I mean in certain end use cases USB can still be handy for transferring files if say for example you have a mechanical hard drive. I've used a thumb drive plenty of times with my DOS 6.22 Socket 7 system , it saves me giving up an expansion slot for a network card or having to remove my HDDs every time I need to put stuff on them. I do use PS/2 devices for my keyboard and mouse though.
USB is a convenient way to transfer files, I agree. There is one benefit I like using wireless USB keyboards and mice: less cables
Since I change my setups a lot, it does help me keep my desk clean. But for benchmarks or long term use, it's probably better to stick with PS/2.
USB is very convenient for the end user but not so much for the developers. It requires entire protocol stack similar to that of bluetooth or networking. In "legacy" mode, simplified USB stack and legacy device emulation is implemented in the BIOS and runs in SMM (System Management Mode). Long story short, it's quite a lot of code that steals CPU cycles from the OS. Switching to and back from SMM takes time. And since it is using CPU caches, just like any other code, it causes even more performance loss. Not to mention that it is pretty much impossible to predict when SMM is going to kick in. Depending on how good or bad the implementation is, it might be pretty much invisible to the OS and the user or break everything entirely, including completely crashing the system. Updating the BIOS or changing the mainboard might change things by a lot.
I have seen SMM mentioned many times now in the comments. I may put the focus on it in the next video as "polling" may not have been the correct terminology.
Lately I've been trying out Settlers III for the first time as I've never had an opportunity to play the full game before, only a demo version back in the late 90s. Imagine my bafflement when I found out that the game freezes completely whenever I move the mouse around. As soon as I leave it still everything is fine again. How can I even play a real time strategy that way?? Surprisingly no such issue occurs when I use the trackpoint (the computer is an IBM ThinkPad T42 laptop with a Pentium M 1700 MHz; not all that retro, merely 18 years old). I have no idea what produces such weird symptoms. I wanted to give a PS/2 mouse a try and see if it works fine with the game but after closer inspection what I've hoped to be a PS/2 port turned out to be an S-video out so no cigar ;)
I've been capturing footage of Settlers III - for a new series about copy protections back then.
I played it on a Pentium 200 with 96 MB memory. I even used a USB keyboard and mouse. It wasn't running well, but good enough to make it to Romans scenario 7.
I have not encountered any such issues you describe. However, I once experienced a weird behavior in Jedi Academy (I think it was that game). I could play the game until a certain point, then, the frame rate went to 1 frame every 5 seconds. This was on a newer Intel Celeron N3450. I tried different video drivers and all kinds of other remedies. But I don't think it was related to input
Would this penalty be incurred even if you use USB mice and keyboards with PS2 adapters?
I remember USB being so bad in the 90's that when I saw computers in 2006 that only had USB I'm like WTF!?!?!? Then I realized it wasn't bad anymore. Now I'm like.....why does my 2021 motherboard have PS2? lol
@BuB ich hätte da evtl. eine alternative Lösung für das USB-Problem. Ich habe Anfang diesen Jahres auch damit angefangen mich wieder mit alter Hardware zu beschäftigen. Mein "erster" Classic PC war ein HP Vectra 486. Leider stand ich schnell vor dem Problem, dass die Tastatur nicht komplett funktionierte (ein paar wenige tasten reagierten nicht) und ich dachte mir - das muss doch auch mit einer aktuellen Tastatur lösbar sein. Ich habe also einen kleinen USB auf ps2 Adapter entworfen. Funktionen aktuell: USB Maus oder Tastatur wird automatisch erkannt und auf PS2 Protokoll umgesetzt. Per Schalter lässt sich zwischen XT und AT Mode wählen (XT aber lediglich Tastatur Support).
Falls du Interesse hast, könnte ich dir einen Adapter zukommen lassen sobald ich meine letzten Optimierungen mit eingebaut habe - vielleicht schaffe ich es auch noch bis dahin ein Gehäuse zu entwerfen ;).
Ansonsten tolle Videos! Schaue mir gern deinen Content an und dafür, dass du in der Elektrotechnik lt. deiner Aussage "ein Laie" bist, sind deine Ergebnisse recht gut finde ich (bin gelernter Elektroniker) :).
Hallo! Das hört sich ja nach einem super Projekt an. Ich bin sehr an solchen Lösungen interessiert und würde gern ein solches Projekt/Produkt testen.
Weiteres können wir gerne per E-Mail besprechen (bitsundbolts at Gmail Punkt com).
Würde mich freuen!
@@bitsundboltsFalls die Mail im Spamordner gelandet sein sollte - E-Mail ist raus 😉.
Tatsächlich war sie dort. Ich melde mich! Vielen Dank!
1-3% seems lower than I would have expected. You wouldn't think that would be visible, but I remember it being visible to the naked eye (frame rates dropping).
I may have to run different benchmarks to get a better understanding of how bad the performance loss could be - especially games under windows. I'll also try to look at different USB chipsets and USB versions next.
@@bitsundbolts It may have actually been related to windows. I don't remember
No cyrix cpus tested?
I don't have one yet...
You make a good point, and it is a great video, but if you are afraid of 1-3% speed losses, then maybe playing on a 20 year old computer is not for you? Whatever era you are trying to "cosplay", you will always be able to get top-of-the-line components that were faster than what you had access to in that era at the same price or cheaper (in respect to whatever the era in time is that invokes your melancholy response)
What are the X and Y axis-es on your graphs? Thank you.
Memory/cache size and MB/s
I felt performance hit on my computers all the time and never got it why, so I wasn't wrong all these years.
early implementationsv of usb was terrible i managed to get the usb breakout for an asus super socket 7 mobo with a k6-2 475 and the beta bios and the hacked together driver. its caused so much instability it only ever semi worked on xp sp2 lol
I did some tests and there’s a 20% performance penalty when using usb mouse during games compared to a ps2 mouse
Do you remember what games you used?
....my main question here - Who is tasked with the actual polling? Is that task mandated to the main CPU or does the USB controller contain circuitry for it...? I think, from the problem itself, that the CPU is included thus has a 'periodic leave' from its normal operation, hence the periodic dips in those graphs. But what happened with P3-1000? Is it just fast enough to make those interrupts invisible? Did it run in a different motherboard than the P2? What does that board show with a slower CPU, like a Mendocino Celeron? Maybe its USB controller is more advanced and took over from the CPU at last....? So, not one question after all, I'm full of questions.
Yeah, you just created content for another video right there. The PII was in a Gigabyte slot 1 motherboard. The PIII was in the Asus P3B-F using a slotket adapter.
It could be that the difference in USB controllers makes a difference. I do have other slot 1 CPUs to dig even deeper. Maybe a good test while keeping the motherboard the same
The BIOS includes a emulation. On USB1 odr USB 2, the CPU has to handle the polling. Additionally it will take you some konventinal Memory, in most cases. I saw something beetweeen 8 or 16KiB used. USB Input devices should be avoided in general.
@@huberthans4312 8-16k? That's an inconveniently big chunk. I'll look into it (too) when I have some more modern piece on my test bench. One more reason to leave USB alone...
In this era, i have microlag when i review the stage in Colin McRae Rally 2001. I optimizing windows, not work, i rearranged the device IRQ-s, with exchange cards in the slots, and not work, but a little better. I have shared IRQ on the USB controller, and not able to change, but i only use USB when printing. Then disable USB in the BIOS, now the fps better 45 vs 54. This time i use Teac 8x CDRW, but this only PIO4 multiword DMA mode. It work super, and not cause any problem. I bought a faster Phillips 16x CDRW, and this UDMA66 capable, now the microlag not appear in the game. I later found a solution, a windows setting autoinsertcdnotification to disable and it work, no microlag with the old CDRW connected, but the new UDMA device never appear the microlag. I making test, when i set all device IRQ-s to 9, this cause -10fps in the game.
You spent a lot of time debugging the issue! Great that you found a solution! Issues like those are so difficult to resolve!
@@bitsundbolts I testing computers in this era, with Quake2 crusher.dm2 demo in loop for power and stability. I upgraded my cpu from Thunderbird 850 to Athlon XP1900. The fps jump from 30-32 to 38-40 in Quake 2 demo. I not interested. My little broder using Impulse Tracker, and we have AWE64 ISA card, and sound good, and very good compatibility, for every program. A little later i have little money, and bought a SB Live!, just for fun. It is compatible also. When updgrading the machine i select motherboard with ISA slot, to using AWE64, this time is not planned to change the sound card. Later coming the 3DMark with 3D sound test, and wow, this is good. I think, test the Quake with A3D setting, and mmm. Now test the Crusher.dm2 demo, with normal stereo, and with this sound card the fps jump from 38-40 to 55-58. Now i think, the modern computer using ISA device is a big brake, and if not enogh knoledge, the PCI also braking a little bit.
Yes, I have read it many times that ISA sound cards may slow down a system. Depending on what I am measuring, I test without a soundcard installed.
I would guess that when the serial mouse is being used that the cpu needs some cycles for it but not when the mouse driver is not loaded.
Interesting thought. I just wonder if it will be visible in the cache benchmarks. Might be something interesting to look at in another follow-up! Thanks!
Have you tried disabling the USB controller completely? I've heard that just having it enabled drops performance...
Yes, I have tested this, but I did not notice any difference. Having no USB input device connected is equal to disabling USB in the BIOS. At least this was the behavior of the hardware and benchmarks I used.
ofc USB didn't affect pentium 3. Pentium 3 was a big jump in performance in Coopermine and even more in Tualatin cores. Pentium 3 are very fast machines, i love them.
08:30 The memory throughput is about 60MBps lower when USB is connected. Probably not enough to be noticeable
Yes, I have noticed this large drop as well. As you said, probably not enough to notice a difference in the benchmarks I used. Maybe there are others where it is noticeable again.
i know retro pc's are a thing but ....this is like saying you can over clock a raspberry pi 5 . ANY of these issues can be resolved by UPGRADING to a faster computer with more modern i/o
I wouldn't mind seeing a benchmark/actual time difference going from HDD to CF to SSD on old PCs.
I have some old HDD arriving soon. If they work, I will make some HDD videos in the coming months as well. And a great idea - my retro systems are snappy and fast - and that is probably because of the ide-to-sd card adapter. I already feel the pain getting into Windows when using one of those old HDD!
I have several retro computers that have usb and none of them have any issues with pendrives or keyboard and mouse. I just use usb in order to use pendrives but I ran tests with a wireless keyboard and mouse and everything was ok. No change in performance.
That's interesting. 3 of the 4 boards I have tested in this video show the issue visually in SpeedSys when an input device like a keyboard is connected via USB. Could you share the hardware you're using?
Nice video. But you still don't include the USB controller type to comparison. Via chiset is using uhci controller which is the worst possible one. Ohci controllers are much better and will result in less cpu usage.
I'm in the process to source different PCI USB expansion cards (USB 1.1 and USB 2.0). I hope to be able to reproduce the experience many commenters share.
I wonder, if disabling 'legacy device emulation' in BIOS would speed things up under windows. (But this will entirely disable USB input under DOS)
I have to look into this more in depth. I don't think that the two socket 7 motherboards have that option in the BIOS. They only allow to enable or disable USB. I may test this with PIII and maybe even a P4 system
Jakieś 25 lat temu, zauważyłem spadek wydajności na BX440 po dodaniu karty USB
are the results the same with all USB HID devices?
At the moment, I would say yes. I tested wired and wireless HID devices, and they all show this behavior. The issue is not present with USB storage devices like a thumb drive.
Meanwhile my SiS chipset works only at 40% RAM throughput if I dare to enable the onboard VGA graphics XD
Ahh, the pitfalls of modern PCs :)
@@bitsundbolts A "modern" Super Socket7 system LOL
that logitech wireless adapter might have more processing power than the host cpu
So only usb 1 is an issue because it doesn’t have its own hardware to use.
So usb 3 should be used cause it has its own chips to process data independently of the system itself?
would using an USB device connected to a PS2 port using an adaptor affect performance?
Probably not. If the USB device is capable of communicating via an adapter to the PS/2 bus, then it implements the correct protocols. In that case, it would behave like an ordinary PS/2 device
Ah phew most of mine are from the later 90s,
It would have been interesting to see, if disabling the USB controller from BIOS has any effect. ;)
I did disable the USB controller in one test which I did not include in the video. It is identical to no input device connected to the ports. So, disabling the USB feature in the BIOS does not impact the system.
The start wasn’t exactly great, made it sound USB came and replaced FireWire, were two different types of connections back in the days, FireWire for high speed data transfer and USB for slower like printers mouse keyboards and FireWire for video external storage
Overtime they increased the USB data transfer speed and eventually FireWire became less used due to certain factors like majority of PCs just had USB unlike Mac PCs and license
Yes, you're right. But there was a problem with FireWire - not many PCs had it built in. I remember that I required an expansion card to have FireWire available. Later, a FireWire port was more common to be on the motherboard.
You are correct that FireWire was the way to go for storage devices until later USB revisions.
Interesting.
Well retro pc will probably support usb 1.0 and 1.1 better since thats what usb existed in the 90s
Before usb you had Atari sio, literally usb before usb, native plug and play, supplies power, serial data, hardware handshakes, supports hubs and daisy chaining, and hardware controlled transfer speed up to 128kbyte/sec
So you can write all 512k ram in atari computer (if u have ram upgrade past xe and 64k) in about 5 seconds
And atleast my keyboard is built in on my atari
Always choose the blue pill 😜