I love the methodical way you approach troubleshooting this board. Most people would have just discarded it as a poor design. I am always glad to see a new video of yours in my recommendations.
Great video and awesome board. I have the same board and it is as you described unstable as hell. I will definitely put the capacitors there as well 👍🏻. And wow, I am very surprised about your frame rates for the #isadoom25fps challenge! You outperformed me already before I released my video. how cool is that 😂. Thanks for sharing your great experience and superb content with us. Cheers, Peter
Thank you Peter! I don't think, this is it. I have a feeling it's possible to get even more out of it. I'd be curious to know, if your board runs at 40MHz with a 386. May be there is a defect on my board, which prevents it from delivering 80MHz on CLK2.
You two have amazing content and have inspired me to push my AMD K6-2 to the max. I know it's much newer than what you work with but it was my first PC in 2000 as a high school present and very important to me
I am eagerly waiting for the 200th part, where you make this board compatible with Ryzen 7x00 and Intel 13x00 CPUs and latest GPU's through a heavily modified and overclocked ISA bus.
Excellent tinkering in order to discover that there is a problem with the voltage and adding more decoupling capacitors! Also spending hours to tune and bench the system! That is our love for vintage stuff. SimCity2000 is a game that indeed takes advantage of a fpu. SC2000 is almost unplayable with a 486DLC/40. Minimum CPU is a 486DX-40. Maybe a fpu helps.
So glad to see part 2 come so quickly, really looking forward to part 3. I am happy you are able to produce such excellent content. BTW the capacitor implementation was genious.
Great video! The capacitor tweaks were brilliant. I have a VIA based FIC 486 board with some odd stability issues that I will look more into too. Looking forward to trying my hand at the #isadoom25fps challenge 🙂
For HF ripple you need ceramic caps with low ESR. The electrolitic one has bad ESR and actually a high serial inductance. To block HF ripple you need a lot of small MLCC caps in parallel close to CPU.
i really wished you would make so much money with yt so that you would release a video every week and quit your day job . this is how much i enjoy and like your videos. you and vswitchzero are complete Sherlocks when it comes to repairing, i just love it. keep it up
This series is being gold for me. I have my very first PC, a 486 dx66 vlb board very similar to that one in a very unstable state after several decades on storage. I moved back to ISA cards both the graphics card (also a Cirrus Logic) and the IDE card (promise cache vl), but it still was randomly crashing and refusing to boot. I will take a look at the ripple as that might be a cause. Maybe old caps could be showing good values in capacitance and ESR but doing poorly in filtering after those decades. Thanks a lot!
Good catch on the power issues. "You are what you eat", they say, and that is very true for computer hardware. I read stories about hardware became much more overclockable (as expected) and producing less heat (not so expected) with cleaner power rails. Also on some "PC Chips" boards they used cheap aluminum electrolytics for filtering (no tants, almost no ceramics), which ofcourse dried out after 25 years of service, and those boards are sometimes very unstable without a full board recap.
That ripple problem is very interesting. I recently started trying to measure power supply ripple, so I know what to recap and what not, but there seems to be ripple created on the board itself. Oh dear...
You are an outstanding tech. I really enjoy these videos. Specially since I transitioned from a 386 to a 486 during that period of time. And yes, on that 386 I played Doom in a tiny window because that's how it had to be done.
The polling clock rate is generated from that smaller crystal onboard and used for the PS1/keyboard controller and if you had an onboard floppy controller. Typically.
outstanding results! I hope changing the dram chips on the isa video fixes the artifacts, but I think it might not fix it. however, pushing the isa so far is really cool! thank you for this video!
WOW!!! that was a GREAT RESULT!! I didn't even think people overclocked back in the 486's hayday (well, not to get performance.. only for compatibility) What a GREAT CHALLENGE!!! I cannot wait for the next part!!! :D
That's true. Where luck decreases the more experienced you become, since you develop a feeling where to search, if something goes wrong. Often you have no idea what exactly it is, but sometimes you can exclude what it is definitely not. I'm a software developer since 30 years and it's the same there, but I'm by far more confident in that, than in electronic engineering :)
@@necro_ware Yeah. My job involves a lot of troubleshooting, too (and I am a hobby coder as well.) So I'm very familiar with procedural troubleshooting. I always say it's a process of trying to divide the problem domain in half over and over again until you narrow it down. But, sometimes, you just run smack into something by accident, and happen to see a piece of the puzzle. I don't mean that to detract from the skill involved -- it takes skill to be successful. Just kind of marveling at how it is sometimes only fate that leads us to an answer. Or at least, it gets us there more quickly. Just think about how long you could've gone before thinking to check the 5V rail ripple, without having noticed a difference between two PSUs. :-)
Again, great video! I love watching how you come to conclusions and track down each issue step by step. It's like watching a detective show on TV, only with PCBs :) awesome content, as always!
I would love to create a custom case for this system, with every jumper connected to an external switch so that you could easily, quickly, and conveniently change the motherboard settings on the fly. This, coupled with the insane compatibility of the system's board, could make this into a really neat test bench / benchmark / diagnostic system.
Another awesome video. I really enjoyed the deep dive into the cache termination and the ripple fix. Amazing troubleshooting skills got you a rock solid, unique, and well performing board.
I love your videos! I always wanted to populate missing ICs in electronics to get some extra functionality out of them and watching you do it and have good results makes me want to give it a shot to see if it works. Keep up the great work!
@@necro_ware I started my computing career using all these types of boards and processors. I remember how excited I used to get when something new came out. Like VLB or the next biggest processor that I couldn’t afford 🤠
Without the stabilization work, it was not possible to overclock or tweak the board at all. It was more or less stable only on the worst BIOS settings.
absolutely need stable power, years ago my friend was overclocking his 1.6ghz amd althon64 and couldn't get much more then 1.8ghz out of it, months later i put his cpu in my machine and managed to get 2.2ghz. he was using some generic discount psu, while i was using a higher rated psu. even managed to get a better overclock on his ati radeon 9550.
Nicely done! I guess I know now what to look for when I have unstable boards ;) It would have been really nice to have seen video memory throughput here though. Benchmarks like Doom are CPU limited, and having the raw MB/s to/from video memory would help to understand if or if not VLB would make a difference for 386 machines. Have a good bench I never published. Can send it to you. About the clock speed: I saw that there's a position where you could solder in a crystal beneath the CPU. Maybe I didn't catch it, but could the hack be to insert an 80 Mhz crystal there and jumper the system to 486 and 40 MHz bus?
Yes, this is what I meant in the video as I said that an own clock oscillator or generator can be added. But I would have to reverse engineer the traces to know definitely if it is the way to go. In regards of throughput, I made those measurements, just didn't prepare them nicely, because they are more or less the same as what Peter from CPU Galaxy presented once in his video. If the number would deviate, I'd publish them in this video.
@@necro_ware I guess I was a little tired already, I often watch those videos at night... You're right, he benched it already. I guess I didn't notice (saw his video some time ago, too!) Anyway, thanks for sharing something about these rare boards!
That KTS IC is the PLL clock generator. Just check yhe 14MHz crystal pins to see that are connected to it. The control pins are driven by the chipset instead of setting them manually with jumpers. Since it generates correctly other frequencies, I suspect a bad solder joint or a cracked trace. Also might be a bug and that explains the unpopulated 386 side of the board.
Cool board, Scorp! Super cool content 👌 That Am386DX-40 should run fine with an external 80MHz oscillator indeed. The Speedstar VGA is fast indeed and I love the Tseng ET4000 chip for its speed and compatibility but is isn't the fastest ISA VGA. The Diamond Speedstar 24x with the WD90C31 chipset actually and the Trident TVGA8900D (yes a Trident; no typo!) both beat the Tseng (by a hair ;-) ) Any of those three are excellent choices though for ISA builds.. Indeed regular DooM is not enjoyable on any 386 really. Replace a PGA-132 with a Ti486SXL2-50 and it will be.. On slower machines I run FastDooM which makes it somewhat playable even on my 386SX-16 🙂 My 160MHz 486 (Am5x86) build will probably break the 25FPS challenge as well with a fast ISA card but that's not a really period correct setup but hey it's a 486 ;-)
Thank you! Yes, I have an WD90C31 and 8900D here, they are indeed similar fast as the Tseng ET4000 and I use them regularly too. At stock clock, I don't see any difference in performance, but the SpeedStar is more reliable with the overclocked ISA bus. I don't have the SpeedStar 24x, hence I say, that the SpeedStar is the fastest ISA card which I have. And I don't like the 8900D, because it produces jailbars on the LCD displays. In my opinion the best compromise card is the CL-GD542x based ones. Since they have a better integrated DAC, a very good image even on LCDs, very good games compatibility (ET4000 is not that great, WD90C3x is much as well) and they are nearly as fast as the others. In my own builds I always go with Cirrus Logic.
@@necro_ware Yeah the ramdacs on Trident cards leave much to be desired but since I am allergic to LCD screens, the destroyers of pixel art the less noticeable difference in quality on most analog CRT screens does not really bother me personally. The best external VGA ramdacs I've used are the BrookTree ones and the early gold ceramic Inmos units. They make every CRT shine ;-) The Inmos DAC on my IBM PS/2 Model 30's MCGA, hooked up to a PVM: 256 color 15KHz CRT heaven ;-) In my experience the Trident 8900D and Speedstar 24x (which is Diamond's iteration of the WD90C31) also work great with an overclocked ISA bus in my case. They're both very stable even in 0WS mode and at higher speeds in my builds which are mostly upgraded 386 machines (SXL2, SXLC2 and DLC) and do beat the Tseng ET4000 albeit by a hair.. The Cirrus ISA cards are indeed very compatible but I find them too slow.. I do really like Cirrus' later VLB iteration of the Alpine series chipset (CL-GD5434) on my P5 Pentium (yes a Pentium with VLB) build with its 64-bit memory bus. That card flies! If you're looking for the holy grail of ISA VGA cards: Look for a Diamond Speedstar 64 which has that exact same CL Alpine chip, breathes through a 16-bit straw yet still manages to kick every other ISA card's @ss ;-)
@@pipschannel1222 Unfortunately for CRT you need a lot of space, which I don't have, so I have to live with LCDs, therefore I have two, which can handle 15kHz. Regarding Cirrus Logic, that's strange. In my tests the GD542x cards were on par with et4000 and the others. Only less stable on higher clocks. But I also should mention, that I have 5 different ET4000 cards and they all behave differently when overclocked, so I guess, it's also a matter of a layout and even day of production.
@@necro_ware Yeah there are definitely a lot of differences in board designs of cards using the same chip.. Also BIOS differences can play a big part here, especially in terms of compatibility. One thing is clear about the cards we both mentioned here: For the ISA bus it doesn't get any better as almost all of them manage to fill it up to the brim (which doesn't take a lot) obviously making that old bus the biggest bottleneck of the system. It is however very cool to get the most out of an ISA system. In my experience overclocking the ISA bus is the only way to experience any real speed boosts. The Diamond Speedstar 64 ISA with the Alpine CL-5434 chip is the overall winner in my benchmarks as the ultimate ISA graphics card as it ticks all the boxes: It's very compatible and very fast with old DOS games (it's a Cirrus after all), has an interleaved 64-bit memory interface and advanced GUI acceleration making Win311 and Win95 shine at high res with its 2meg framebuffer and has an excellent internal RAMDAC 👌 The VLB version of this card is also lightning fast and way ahead of its time..
Ok. In interest of the extreme and impractical, quake uses a math co-pro. Will it run on the 386. Yes, I'm sure it'd be an absolute slide show if it did, but will it even load? I recall lots of software that claimed to need a 486, really didn't use the more advanced architecture, but would just run like crap because they really needed the 486 for the speed. I don't hold high hopes that quake uses only 386 instructions, but at least give it a shot to crash. But even more crazy, a creative labs 3d blaster vlb. The only vlb 3d accelerator that I know of, and it uses a pass through like the old monster brand cards. A 386 running your fastest vlb main video card, plus the 3d blaster, with math co-pro and your best settings, and grab as many 3d blaster supported tites to test on it and see what'll actually run. A 386/7 with hardware accelerated 3d would be super ironic and hilarious, even if it's slower than frozen snot. Lol.
Yes, interesting topics indeed. Unfortunately, it is nearly impossible to get a VLB 3D accelerator today. But luckily Atheatos already answered most of the questions regarding Quake on a 386 and FPU. Have fun: th-cam.com/video/KDQsLDN9df8/w-d-xo.html th-cam.com/video/Zvisa_-uEqI/w-d-xo.html
So, here's another interesting question: How fast can you crank that ISA bus and still have a sound card work? Because that (and the HDD controller, it'll profit from the extra bandwidth till it faceplants) are usually the limiting factors regarding ISA bus overclocking. If you could run an Audiodrive at that speed, it'd be one hell of a setup for wringing out every last bit of performance from those late ISA video cards, like the Mach 32/64, the ET4000-W32i, and the GD-5434 (assuming they could handle the fast ISA bus).
heh, at least you had more luck with this. i'm still stuck trying to get any change in behavior out of my intel pci/ed. no vide, but it beeps like its posting, and does the memory check noise. also gives the memory error if i pull the ram out. all connections trace out valid, and it doesn't matter which video card i put in it, isa, pci, no response. everything works fine in other machines, including the cpus, ram and video cards. no voltage drop beyond the expected few millivolts. the only odd thing i can find is the fact that the cache datalines appear to have 3 signals present on each. someone tried to tell me to set up my scope right, except.. every other line on the board doesn't have that phenomenon. hooking it up to my function gen and cranking it up to where the signal freq from the cache lines is yields no multi-signal patterns using identical settings. the rst line is behaving as it should, and i'm getting clk and voltage to the cpu socket. i really don't get where the issue is, and i think its close to going back in the box for the forseeable future.
i assume polling rate refers to the keyboard polling... but then again i thought AT/PS2 keyboards use Interrupts and aren't polled. either way good video, this board is pretty damn interesting. and it makes me kinda jealous that my own only 386 system i have has a short on the 5V rail and i don't know why :(
Yes, I also thought about the keyboard, but didn't come to investigate. The keyboard controller is actually polling, it handles the keyboard input and sends an interrupt if something interesting happens. In regards of short, it's almost always a tantalum capacitor. If checked that and they all are good, you could use voltage injection to find the culprit. I showed it once here for a more modern hardware, but this can be of course done for the old hardware as well: th-cam.com/video/SFxOFyp6uFU/w-d-xo.html
@@necro_ware oh, in that case maybe it's as simple as it changing the clock speed of the keyboard controller, causing it to poll faster. or it just sends some kind of command... anyways, i also thought it was a dead tantulum on my system, but even after replacing all of them with new ones the short still persists. and sadly i don't have a bench power supply for voltage injection.
Ist the resistor Network RP9 in the near of the srams in a good state? Ob the Video Looks Like bend. I saw it in the Last Video. But i forgot to write. You find it. You are a very good technican
I get 12.1 fps from timedemo using fdoom13h with high settings using my 386DX 40 MHz and ET4000 running in ISA with CPUCLK/2. The 13h variant is quite a bit faster than the regular. The game is surprisingly playable at least on a quick glance.
I noticed that there are settings for the 486DX/50 in the mainboard manual - this theoretically implies, if you could run ISA at CLK/4 that you could have a 12.5MHz AT-bus, and with an Am5x86-P75 or similar AMD Enhanced 486, a theoretical 200MHz CPU clock, all other things being equal. Do you think this board would be up to that?
I found WD90C31 was slighly faster than ET4000 on ISA bus (Doom) but WD has Commander Keen scrolling bug, and VGA screen fade flashing bug also, so not 100% VGA compatible maybe 99.9%. Foiund ET4000 was perfect.
Just a quick comment: the vram latency needs to be also changed in the vbios to set the refresh timers accordingly. Using a "faster" dram without the proper timing has no performance gain
Как же производители материнок любили расставлять компоненты четко по их габаритам, без зазоров... Особенно разъемы АТ vs RAM! На двушке недавно встретилось такое, что жесткие пластмассовые P8-P9 упирались аж в сами чипы памяти 🤦♂Продолжу смотреть этот сериал потом, хорошего понемножку ) А ты FastDoom не пробовал, кстати? Всё-таки поприятнее он идёт, даже на грустных трешках )
TIP: Cirrus 5434 VLB is lot faster than 5428. Damn I love overclocking old stuff, now I'm using Cx5x86/100 @120MHz and it needs at least 4 volts to be stable.
And what happened with the IDE speed? Would the sound card still work if you had used one? Because there's a guy in every country who doesn't care about FPS games and in Spain it just happened to be *me* LOL
IDE make nearly no difference. Overclocked ISA make a lot of problems with sound cards indeed, but it heavily depends on the sound card. I have around 30-40 ISA sound cards in my collection, may be I should make a big stability check one day :)
@@necro_ware I've never used a flash drive of any kind on an IDE ISA adapter, are they anywhere near the 5MB/s that ISA SCSI controllers used to list as the maximum recommended speed? With large disks without smartdrv loaded waiting for DIR to calculate the free space is maddening, samething doing a full format, I don't know if using a flash drive improves anything in that sense, otherwise I don't have any problem with the transfer speed of IDE drives.
In theory a capacitor between the positive and negative pols acts as a low-pass filter. However, in real life the bigger electrolytic cap filters middle frequencies, but is too inert to react to higher frequencies. The small ceramic capacitor has too low capacity to filter middle frequencies, but is much faster and is able to filter higher frequencies. In combination they act as a proper low-pass filter.
This test seems to confirm that the 386 and ISA should indeed be eliminated at the same time, but the ISA has persisted until the 845G chipset motherboard of the Pentium 4 era
Ah the 90s hardware. When you could overclock a part 250% and not crash. Now people get excited about single-digit percent overclocks on their graphics card
I wish you install a pentium overdrive Although I used to change hardware very frequently and also owned a computer shop from 1997 to 2000, I never put my hands on a pentium overdrive. My first pentium was a 120mhz, overclocked at 133, with 512ko cache I changed myself, soon to be replaced by a Cyrix P166+, much faster as long as the fpu was not involved
19:50 ok but The only thing, which I don't like is that you never say that you are using High quality settings. And If you talk about reducing quality, you always talk only about screen size, but never about an feature that in doom's settings we can reduce quality to save fps. Just Image this approach today - when you have Cyberpunk2077 and you are saying - look only 8fps on gtx770, so it is not playable - we can reduce screen size (resolution) but look, it is still not playable. BUT in reality we all can see that you have set everything at the ultra in graphics settings. I just don't get it. What is the point to run a doom, always on highest quality settings, and prove that the lowest cpus (386,486) cannot even handle it. But if we reduce screeen size to size of pingpong ball then we can play. But how about low quality settings? Why you not talking about that?
I love the methodical way you approach troubleshooting this board. Most people would have just discarded it as a poor design. I am always glad to see a new video of yours in my recommendations.
If not for it's design then definitely for it's age and design quirks... sockets for a 386 and a 486?! choose a side buddy
Great video and awesome board. I have the same board and it is as you described unstable as hell. I will definitely put the capacitors there as well 👍🏻. And wow, I am very surprised about your frame rates for the #isadoom25fps challenge! You outperformed me already before I released my video. how cool is that 😂. Thanks for sharing your great experience and superb content with us. Cheers, Peter
Thank you Peter! I don't think, this is it. I have a feeling it's possible to get even more out of it. I'd be curious to know, if your board runs at 40MHz with a 386. May be there is a defect on my board, which prevents it from delivering 80MHz on CLK2.
@@necro_ware I will check that and measure it to verify if it works on my board. 👍🏻
You two have amazing content and have inspired me to push my AMD K6-2 to the max. I know it's much newer than what you work with but it was my first PC in 2000 as a high school present and very important to me
I am eagerly waiting for the 200th part, where you make this board compatible with Ryzen 7x00 and Intel 13x00 CPUs and latest GPU's through a heavily modified and overclocked ISA bus.
That's the spirit! 🤣
Excellent tinkering in order to discover that there is a problem with the voltage and adding more decoupling capacitors! Also spending hours to tune and bench the system! That is our love for vintage stuff. SimCity2000 is a game that indeed takes advantage of a fpu. SC2000 is almost unplayable with a 486DLC/40. Minimum CPU is a 486DX-40. Maybe a fpu helps.
respect your determination on troubleshooting 30year old tech and soft to see small numbers and answer some questions. hats off
So glad to see part 2 come so quickly, really looking forward to part 3. I am happy you are able to produce such excellent content. BTW the capacitor implementation was genious.
Great video! The capacitor tweaks were brilliant. I have a VIA based FIC 486 board with some odd stability issues that I will look more into too. Looking forward to trying my hand at the #isadoom25fps challenge 🙂
For HF ripple you need ceramic caps with low ESR. The electrolitic one has bad ESR and actually a high serial inductance. To block HF ripple you need a lot of small MLCC caps in parallel close to CPU.
That is the kind of mod I have seen hard core overclockers do on their mainboards (TAGG XOC for example) or I'm mistaken?
20:20 thanks for that additional benchmark with coprocessor.
The only channel I still watch
i really wished you would make so much money with yt so that you would release a video every week and quit your day job . this is how much i enjoy and like your videos. you and vswitchzero are complete Sherlocks when it comes to repairing, i just love it. keep it up
LOL, that's a long way to go :) Furthermore, I released less videos last year mostly not because of my work, but because of my volunteer duties.
Nicely done! Finding that resistor pack was the stroke of luck you needed, and it put you onto so much more. I'm looking forward to part 3.
This series is being gold for me. I have my very first PC, a 486 dx66 vlb board very similar to that one in a very unstable state after several decades on storage. I moved back to ISA cards both the graphics card (also a Cirrus Logic) and the IDE card (promise cache vl), but it still was randomly crashing and refusing to boot.
I will take a look at the ripple as that might be a cause. Maybe old caps could be showing good values in capacitance and ESR but doing poorly in filtering after those decades.
Thanks a lot!
Good catch on the power issues. "You are what you eat", they say, and that is very true for computer hardware. I read stories about hardware became much more overclockable (as expected) and producing less heat (not so expected) with cleaner power rails. Also on some "PC Chips" boards they used cheap aluminum electrolytics for filtering (no tants, almost no ceramics), which ofcourse dried out after 25 years of service, and those boards are sometimes very unstable without a full board recap.
Very good stuff. I am seriously impressed with these end results. This gives an extremely good retro system overall.
That ripple problem is very interesting. I recently started trying to measure power supply ripple, so I know what to recap and what not, but there seems to be ripple created on the board itself. Oh dear...
You are an outstanding tech. I really enjoy these videos. Specially since I transitioned from a 386 to a 486 during that period of time. And yes, on that 386 I played Doom in a tiny window because that's how it had to be done.
Great, great, great job! And video as well ;)
I fondly remember playing doom in a tiny window on my 386, and it was still very enjoyable!
The polling clock rate is generated from that smaller crystal onboard and used for the PS1/keyboard controller and if you had an onboard floppy controller. Typically.
Pleasant to watch as always, nice work.
outstanding results! I hope changing the dram chips on the isa video fixes the artifacts, but I think it might not fix it. however, pushing the isa so far is really cool! thank you for this video!
WOW!!! that was a GREAT RESULT!! I didn't even think people overclocked back in the 486's hayday (well, not to get performance.. only for compatibility)
What a GREAT CHALLENGE!!! I cannot wait for the next part!!! :D
Man. This really illustrates how much troubleshooting is often just a series of lucky guesses and dumb luck.
That's true. Where luck decreases the more experienced you become, since you develop a feeling where to search, if something goes wrong. Often you have no idea what exactly it is, but sometimes you can exclude what it is definitely not. I'm a software developer since 30 years and it's the same there, but I'm by far more confident in that, than in electronic engineering :)
@@necro_ware Yeah. My job involves a lot of troubleshooting, too (and I am a hobby coder as well.) So I'm very familiar with procedural troubleshooting. I always say it's a process of trying to divide the problem domain in half over and over again until you narrow it down.
But, sometimes, you just run smack into something by accident, and happen to see a piece of the puzzle. I don't mean that to detract from the skill involved -- it takes skill to be successful. Just kind of marveling at how it is sometimes only fate that leads us to an answer. Or at least, it gets us there more quickly. Just think about how long you could've gone before thinking to check the 5V rail ripple, without having noticed a difference between two PSUs. :-)
Great video, looking forward to seeing more on this.
It is a nice thing about your videos - I still can learn something more from you.
Probably, because I usually don't make videos, just to make videos. I do most of my projects to learn something new.
@@necro_ware Right, dealing with practical problems results in most valuable experience. Thank you for sharing it.
amazing job in catching - and resolving - the excessive ripple on the 5V line!
Great you are back my frend
Sweet video I always learn a lot from your content
Did I enjoy your video ? I loved it ! And now I will play some Doom !
Again, great video! I love watching how you come to conclusions and track down each issue step by step. It's like watching a detective show on TV, only with PCBs :) awesome content, as always!
The DX4 will be able ot stretch it a little further. I look forward to your next video! Thanks Necro!
Fantastic video!!
awesome video, cant wait for the next one!
Very impressed with the FPS you got in Doom. Back then, I couldn't get nearly that much out of my 386DX-40.
I'd love to see the ISA graphics results with a 486 DX-50!
I would love to create a custom case for this system, with every jumper connected to an external switch so that you could easily, quickly, and conveniently change the motherboard settings on the fly. This, coupled with the insane compatibility of the system's board, could make this into a really neat test bench / benchmark / diagnostic system.
❤
More is not necessary to say!
love this video, very well made :)
Love your videos NW!
one of the coolest motherboards iv seen on youtube
Another awesome video. I really enjoyed the deep dive into the cache termination and the ripple fix. Amazing troubleshooting skills got you a rock solid, unique, and well performing board.
Great video! Getting an external clock on their might allowed you to push it even further.
Loving your videos! Very therapeutic but also learning a lot about this mid-90s transitional period of standards.
Amazing job!!
Perfect! Waiting for third part.
I love your videos! I always wanted to populate missing ICs in electronics to get some extra functionality out of them and watching you do it and have good results makes me want to give it a shot to see if it works. Keep up the great work!
Wow excellent work knowing where to add not one but two capacitors. I just watch and think how do you learn that!
Nice job.
I read a lot about power supplies as I did my MMX-VRM project. There such things like ripple are a very important topic.
@@necro_ware
I started my computing career using all these types of boards and processors. I remember how excited I used to get when something new came out.
Like VLB or the next biggest processor that I couldn’t afford 🤠
Phenomenal !!!! Thank you for sharing ...
Brilliant
He is a mad man!
17:23 Where do you have that simm storage rack from??
I drew and printed them. May be I should upload the models somewhere, if you need them.
Awesome
I wonder if all the work you did to stabilize power is the key to such a great overclock.
Without the stabilization work, it was not possible to overclock or tweak the board at all. It was more or less stable only on the worst BIOS settings.
absolutely need stable power, years ago my friend was overclocking his 1.6ghz amd althon64 and couldn't get much more then 1.8ghz out of it, months later i put his cpu in my machine and managed to get 2.2ghz. he was using some generic discount psu, while i was using a higher rated psu. even managed to get a better overclock on his ati radeon 9550.
Love to see the ISA card get some sockets for the memory, make it easier to test if one chip is bad.
Just a note - doom on the 3do used a similar window size :-)
It didn’t sell well and everyone hated it. So I think this supports your argument 🎉
Nicely done! I guess I know now what to look for when I have unstable boards ;)
It would have been really nice to have seen video memory throughput here though. Benchmarks like Doom are CPU limited, and having the raw MB/s to/from video memory would help to understand if or if not VLB would make a difference for 386 machines. Have a good bench I never published. Can send it to you.
About the clock speed: I saw that there's a position where you could solder in a crystal beneath the CPU. Maybe I didn't catch it, but could the hack be to insert an 80 Mhz crystal there and jumper the system to 486 and 40 MHz bus?
Yes, this is what I meant in the video as I said that an own clock oscillator or generator can be added. But I would have to reverse engineer the traces to know definitely if it is the way to go. In regards of throughput, I made those measurements, just didn't prepare them nicely, because they are more or less the same as what Peter from CPU Galaxy presented once in his video. If the number would deviate, I'd publish them in this video.
@@necro_ware I guess I was a little tired already, I often watch those videos at night...
You're right, he benched it already. I guess I didn't notice (saw his video some time ago, too!)
Anyway, thanks for sharing something about these rare boards!
That KTS IC is the PLL clock generator. Just check yhe 14MHz crystal pins to see that are connected to it. The control pins are driven by the chipset instead of setting them manually with jumpers.
Since it generates correctly other frequencies, I suspect a bad solder joint or a cracked trace. Also might be a bug and that explains the unpopulated 386 side of the board.
3:21 this is very similar to DDR/SDR, in ddr you get 2 "ticks" or 2 clocks of data per cycle, compared to sdr
I am still SO MAD I can't buy one of these MB. I haven't been this excited for a 386 since they launched!
Would love to see that board comparing a Cyrix 486DLC vs a real 486 at the same clock :)
Would be interesting to see howuch slower it is :)
If I find a way to convince it to output CLK2 at 80MHz, I'll do such tests.
Note that the Cirrus Logic CL-GD542x cards only have 16bit I/O though it is a VLB card.
Awesome!
The Ceramic capacitor reminds me Back to the future 3 with Doc's "fix" for the Delorian, but in reverse ;)
Cool board, Scorp! Super cool content 👌
That Am386DX-40 should run fine with an external 80MHz oscillator indeed.
The Speedstar VGA is fast indeed and I love the Tseng ET4000 chip for its speed and compatibility but is isn't the fastest ISA VGA. The Diamond Speedstar 24x with the WD90C31 chipset actually and the Trident TVGA8900D (yes a Trident; no typo!) both beat the Tseng (by a hair ;-) ) Any of those three are excellent choices though for ISA builds..
Indeed regular DooM is not enjoyable on any 386 really. Replace a PGA-132 with a Ti486SXL2-50 and it will be.. On slower machines I run FastDooM which makes it somewhat playable even on my 386SX-16 🙂
My 160MHz 486 (Am5x86) build will probably break the 25FPS challenge as well with a fast ISA card but that's not a really period correct setup but hey it's a 486 ;-)
Thank you! Yes, I have an WD90C31 and 8900D here, they are indeed similar fast as the Tseng ET4000 and I use them regularly too. At stock clock, I don't see any difference in performance, but the SpeedStar is more reliable with the overclocked ISA bus. I don't have the SpeedStar 24x, hence I say, that the SpeedStar is the fastest ISA card which I have. And I don't like the 8900D, because it produces jailbars on the LCD displays. In my opinion the best compromise card is the CL-GD542x based ones. Since they have a better integrated DAC, a very good image even on LCDs, very good games compatibility (ET4000 is not that great, WD90C3x is much as well) and they are nearly as fast as the others. In my own builds I always go with Cirrus Logic.
@@necro_ware Yeah the ramdacs on Trident cards leave much to be desired but since I am allergic to LCD screens, the destroyers of pixel art the less noticeable difference in quality on most analog CRT screens does not really bother me personally. The best external VGA ramdacs I've used are the BrookTree ones and the early gold ceramic Inmos units. They make every CRT shine ;-)
The Inmos DAC on my IBM PS/2 Model 30's MCGA, hooked up to a PVM: 256 color 15KHz CRT heaven ;-)
In my experience the Trident 8900D and Speedstar 24x (which is Diamond's iteration of the WD90C31) also work great with an overclocked ISA bus in my case. They're both very stable even in 0WS mode and at higher speeds in my builds which are mostly upgraded 386 machines (SXL2, SXLC2 and DLC) and do beat the Tseng ET4000 albeit by a hair..
The Cirrus ISA cards are indeed very compatible but I find them too slow..
I do really like Cirrus' later VLB iteration of the Alpine series chipset (CL-GD5434) on my P5 Pentium (yes a Pentium with VLB) build with its 64-bit memory bus. That card flies!
If you're looking for the holy grail of ISA VGA cards: Look for a Diamond Speedstar 64 which has that exact same CL Alpine chip, breathes through a 16-bit straw yet still manages to kick every other ISA card's @ss ;-)
@@pipschannel1222 Unfortunately for CRT you need a lot of space, which I don't have, so I have to live with LCDs, therefore I have two, which can handle 15kHz. Regarding Cirrus Logic, that's strange. In my tests the GD542x cards were on par with et4000 and the others. Only less stable on higher clocks. But I also should mention, that I have 5 different ET4000 cards and they all behave differently when overclocked, so I guess, it's also a matter of a layout and even day of production.
@@necro_ware Yeah there are definitely a lot of differences in board designs of cards using the same chip.. Also BIOS differences can play a big part here, especially in terms of compatibility. One thing is clear about the cards we both mentioned here: For the ISA bus it doesn't get any better as almost all of them manage to fill it up to the brim (which doesn't take a lot) obviously making that old bus the biggest bottleneck of the system. It is however very cool to get the most out of an ISA system. In my experience overclocking the ISA bus is the only way to experience any real speed boosts.
The Diamond Speedstar 64 ISA with the Alpine CL-5434 chip is the overall winner in my benchmarks as the ultimate ISA graphics card as it ticks all the boxes: It's very compatible and very fast with old DOS games (it's a Cirrus after all), has an interleaved 64-bit memory interface and advanced GUI acceleration making Win311 and Win95 shine at high res with its 2meg framebuffer and has an excellent internal RAMDAC 👌 The VLB version of this card is also lightning fast and way ahead of its time..
Ok. In interest of the extreme and impractical, quake uses a math co-pro. Will it run on the 386. Yes, I'm sure it'd be an absolute slide show if it did, but will it even load? I recall lots of software that claimed to need a 486, really didn't use the more advanced architecture, but would just run like crap because they really needed the 486 for the speed. I don't hold high hopes that quake uses only 386 instructions, but at least give it a shot to crash.
But even more crazy, a creative labs 3d blaster vlb. The only vlb 3d accelerator that I know of, and it uses a pass through like the old monster brand cards. A 386 running your fastest vlb main video card, plus the 3d blaster, with math co-pro and your best settings, and grab as many 3d blaster supported tites to test on it and see what'll actually run.
A 386/7 with hardware accelerated 3d would be super ironic and hilarious, even if it's slower than frozen snot. Lol.
Yes, interesting topics indeed. Unfortunately, it is nearly impossible to get a VLB 3D accelerator today. But luckily Atheatos already answered most of the questions regarding Quake on a 386 and FPU. Have fun:
th-cam.com/video/KDQsLDN9df8/w-d-xo.html
th-cam.com/video/Zvisa_-uEqI/w-d-xo.html
@@necro_ware sweet talk Clint from lgr to borrow his. Lol. I learned about it's existence from one of his videos.
And I'll check those out.
So, here's another interesting question: How fast can you crank that ISA bus and still have a sound card work? Because that (and the HDD controller, it'll profit from the extra bandwidth till it faceplants) are usually the limiting factors regarding ISA bus overclocking. If you could run an Audiodrive at that speed, it'd be one hell of a setup for wringing out every last bit of performance from those late ISA video cards, like the Mach 32/64, the ET4000-W32i, and the GD-5434 (assuming they could handle the fast ISA bus).
heh, at least you had more luck with this. i'm still stuck trying to get any change in behavior out of my intel pci/ed. no vide, but it beeps like its posting, and does the memory check noise. also gives the memory error if i pull the ram out. all connections trace out valid, and it doesn't matter which video card i put in it, isa, pci, no response. everything works fine in other machines, including the cpus, ram and video cards. no voltage drop beyond the expected few millivolts.
the only odd thing i can find is the fact that the cache datalines appear to have 3 signals present on each. someone tried to tell me to set up my scope right, except.. every other line on the board doesn't have that phenomenon. hooking it up to my function gen and cranking it up to where the signal freq from the cache lines is yields no multi-signal patterns using identical settings.
the rst line is behaving as it should, and i'm getting clk and voltage to the cpu socket.
i really don't get where the issue is, and i think its close to going back in the box for the forseeable future.
i assume polling rate refers to the keyboard polling... but then again i thought AT/PS2 keyboards use Interrupts and aren't polled.
either way good video, this board is pretty damn interesting. and it makes me kinda jealous that my own only 386 system i have has a short on the 5V rail and i don't know why :(
Yes, I also thought about the keyboard, but didn't come to investigate. The keyboard controller is actually polling, it handles the keyboard input and sends an interrupt if something interesting happens.
In regards of short, it's almost always a tantalum capacitor. If checked that and they all are good, you could use voltage injection to find the culprit. I showed it once here for a more modern hardware, but this can be of course done for the old hardware as well: th-cam.com/video/SFxOFyp6uFU/w-d-xo.html
@@necro_ware oh, in that case maybe it's as simple as it changing the clock speed of the keyboard controller, causing it to poll faster. or it just sends some kind of command...
anyways, i also thought it was a dead tantulum on my system, but even after replacing all of them with new ones the short still persists. and sadly i don't have a bench power supply for voltage injection.
Ist the resistor Network RP9 in the near of the srams in a good state? Ob the Video Looks Like bend. I saw it in the Last Video. But i forgot to write. You find it. You are a very good technican
I get 12.1 fps from timedemo using fdoom13h with high settings using my 386DX 40 MHz and ET4000 running in ISA with CPUCLK/2. The 13h variant is quite a bit faster than the regular. The game is surprisingly playable at least on a quick glance.
17:14 вот это коллекция. было бы круто увидеть всю вашу коллекцию как нибудь
I noticed that there are settings for the 486DX/50 in the mainboard manual - this theoretically implies, if you could run ISA at CLK/4 that you could have a 12.5MHz AT-bus, and with an Am5x86-P75 or similar AMD Enhanced 486, a theoretical 200MHz CPU clock, all other things being equal. Do you think this board would be up to that?
I found WD90C31 was slighly faster than ET4000 on ISA bus (Doom) but WD has Commander Keen scrolling bug, and VGA screen fade flashing bug also, so not 100% VGA compatible maybe 99.9%. Foiund ET4000 was perfect.
10:14 Почему не используете розе?
cool
I guess it really is just down to the 386 being that slow, even with VLB it just doesn't have the HP to run DooM above half screen
Just a quick comment: the vram latency needs to be also changed in the vbios to set the refresh timers accordingly. Using a "faster" dram without the proper timing has no performance gain
It would be interesting to See the diffrence between a 486SLC and an actual 486.
Dieser Moment, wenn noch vor dem Jaulen der Daumen-hoch bunt blinkt und dann schwarz wird.
When are we going to try for VLB>PCI bridge chip?
I want to see that video card modification. That's awesome
That's actually just about to change the memory ICs, four of which are even already socketed. I'm not sure if that is very exciting :)
It is to me. I had never thought of that before so it's a new idea to me.
What would the cx486DLC do in this board with VLB. ??
Как же производители материнок любили расставлять компоненты четко по их габаритам, без зазоров... Особенно разъемы АТ vs RAM! На двушке недавно встретилось такое, что жесткие пластмассовые P8-P9 упирались аж в сами чипы памяти 🤦♂Продолжу смотреть этот сериал потом, хорошего понемножку ) А ты FastDoom не пробовал, кстати? Всё-таки поприятнее он идёт, даже на грустных трешках )
Я же его в #doscember22 видео рассматривал. На этой платформе запускал тоже, но на видео не снимал.
@@necro_ware Ааа ясно ) Я doscember на сладкое оставил )
@@RetroPeKar Уже следующий скоро ;)
TIP: Cirrus 5434 VLB is lot faster than 5428. Damn I love overclocking old stuff, now I'm using Cx5x86/100 @120MHz and it needs at least 4 volts to be stable.
And what happened with the IDE speed? Would the sound card still work if you had used one?
Because there's a guy in every country who doesn't care about FPS games and in Spain it just happened to be *me* LOL
IDE make nearly no difference. Overclocked ISA make a lot of problems with sound cards indeed, but it heavily depends on the sound card. I have around 30-40 ISA sound cards in my collection, may be I should make a big stability check one day :)
@@necro_ware I've never used a flash drive of any kind on an IDE ISA adapter, are they anywhere near the 5MB/s that ISA SCSI controllers used to list as the maximum recommended speed?
With large disks without smartdrv loaded waiting for DIR to calculate the free space is maddening, samething doing a full format, I don't know if using a flash drive improves anything in that sense, otherwise I don't have any problem with the transfer speed of IDE drives.
why does 1000uf and 0.1uf in parallel remove so much ripple?
In theory a capacitor between the positive and negative pols acts as a low-pass filter. However, in real life the bigger electrolytic cap filters middle frequencies, but is too inert to react to higher frequencies. The small ceramic capacitor has too low capacity to filter middle frequencies, but is much faster and is able to filter higher frequencies. In combination they act as a proper low-pass filter.
I have owned the Diamond Speedstar 64 ISA, which arguably is faster but severely hampered by the ISA bus.
This test seems to confirm that the 386 and ISA should indeed be eliminated at the same time, but the ISA has persisted until the 845G chipset motherboard of the Pentium 4 era
There is nothing wrong with ISA and it still has uses in industrial settings (PC/104).
Eu tenho dito com k6 mas consigo ítalos wo 98
why are you using the original battery
Because this is part 2 of the video ;)
Ah the 90s hardware. When you could overclock a part 250% and not crash. Now people get excited about single-digit percent overclocks on their graphics card
👍
I got 29.8fps in doom with a trident vga isa on a k5
I wish you install a pentium overdrive Although I used to change hardware very frequently and also owned a computer shop from 1997 to 2000, I never put my hands on a pentium overdrive. My first pentium was a 120mhz, overclocked at 133, with 512ko cache I changed myself, soon to be replaced by a Cyrix P166+, much faster as long as the fpu was not involved
Well excuse me while I go and overclock my ISA bus...
I think its extrapolate, not interpolate in this case.
1:30 break out the bodge wires...
19:50 ok but The only thing, which I don't like is that you never say that you are using High quality settings. And If you talk about reducing quality, you always talk only about screen size, but never about an feature that in doom's settings we can reduce quality to save fps.
Just Image this approach today - when you have Cyberpunk2077 and you are saying - look only 8fps on gtx770, so it is not playable - we can reduce screen size (resolution) but look, it is still not playable. BUT in reality we all can see that you have set everything at the ultra in graphics settings. I just don't get it. What is the point to run a doom, always on highest quality settings, and prove that the lowest cpus (386,486) cannot even handle it. But if we reduce screeen size to size of pingpong ball then we can play. But how about low quality settings? Why you not talking about that?
I told about it in Doscember 2022 video.
As a punishment to that cursed mainboard, remove any good caps and replace them with bulging ones.
😂
:)
@@necro_ware "reverse recap"