What if BlueSCSI, but WAY FASTER?
ฝัง
- เผยแพร่เมื่อ 15 เม.ย. 2022
- Head to squarespace.com/ActionRetro to save 10% off your first purchase of a website or domain using code ActionRetro
I absolutely love the BlueSCSI, an open hardware and open source replacement for your old SCSI hard disks. However, we got some less than stellar benchmarks last time. Today, let's try and get as much speed as possible!
LINKS:
🍎 BlueSCSI: scsi.blue
(Amazon links are affiliated links)
══════════════════════════
💾 For more vintage Apple stuff, please subscribe: th-cam.com/users/ActionRetro?s...
💾 Support these retro computing shenanigans on Patreon! / actionretro
══════════════════════════
Check out my Amazon page with links to my tools, adapters, soldering equipment, camera gear and more: www.amazon.com/shop/actionretro
══════════════════════════
💬 Come talk about old computers on the BitBang Social Mastodon! bitbang.social
══════════════════════════
#Macintosh #Linux #PowerPC - วิทยาศาสตร์และเทคโนโลยี
![[LIVE] : ONE ลุมพินี 69 | คู่เอก "กุหลาบดำ vs นาบิล"](http://i.ytimg.com/vi/PNM6U8e_HMo/mqdefault.jpg)
![ภาพนี้ก็ฮาเหมือนกันนะเนี้ย #2 SS8 [ พากย์นรก MEME.EXE ] | easy boy](http://i.ytimg.com/vi/RMhUgxxD49Q/mqdefault.jpg)
You misread the graphs at the end, check the scale. The SCSI2SD has a max of 2048 while the Blue only has 1024. The Blue is 975 while the SCSI is 1330. The Spinner is ~ the SCSI2SD
Yes, the scale is different. Throughput is quicker, but seek times much slower. Once the disks fill up, I wonder how much real-world use is affected by fragmentation?
Indeed. It's almost like Apple performance graphs are forever doomed to relative performance comparisons and hard to read axes, lol.
Still an interesting video.
64TBW stands for 64 Terrabytes Written. It's an endurance measurement, how long the NAND will last before it begins to fail. Higher is better :)
Which is VERY impressive for a 128gig µSD card
Yeah. That's a VERY impressive TBW for a 128 GB card.
Extending what you said with some additional info. Higher TBW is indeed better, all things being equal, but not all things are equal, the number of levels per cell are inversely correlated with write endurance.
SLC has more write endurance than two level cache, which has more than TLC, which has more than QLC.
So if you want more write endurance (TBW), fewer levels per cell is better. However, there's a caveat, SLC is more expensive than more dense options.
So, if you're trying to buy an SSD that's going to end up with very rarely having new data written or rewritten on it, QLC with lower endurance may actually be the better option for your use case.
For instance, if you're gonna install a bunch of games on it and just leave them there indefinitely, higher endurance isn't gonna do you any good, you've just paid way more for something that doesn't get you anything.
Sounds like it'll take for-fucking-ever for it to fail, which is good
@@DryPaperHammerBro depends on exactly how you're writing to it. If you're writing the entire card every time you flash a new OS, well... 512 times before you exceed it.
Man, it warms my heart so much to see Mark Ruffalo's kid into retro tech.
The kid Mark had when he was five?
Thanks for the followup - Great to see the tips helped you speed it up quite a bit!
Thank you so much for this project.
At 12:26 the y-scale of the two graphs is off by double. The performance of the SCSI2SD was essentially double (1024 vs. 2048 max y-scale)
Is that a Kitboga shirt?
I see you're a man of culture. Kit is a magician.
This channel is awesome and partially responsible for my renewed obsession of acquiring and repairing retro macs that I salivated over in the 1990s but was way to broke to afford at the time. Thank you for covering topics like this. FANTASTIC! Much appreciated.
SCSI2SD is based on (up to) 80MHz CortexM3 MCU with 64KB of RAM, BlueSCSI is based on (up to) 72MHz CortexM3 MCU with 20KB of RAM. Speed of BlueSCSI drops suddenly at 32k (sequential or random doesn't matter) when SCSI2SD makes much more consistent results. PSoC5LP doesn't offer much over STM32F103 we find in bluepill. It has advantage with programmable logic within I/O but that hardly explains speed difference (or rather - speed drop of BlueSCSI). My guess was and still is - 20KB of RAM is suboptimal. I wonder if black pill wouldn't make a huge difference. It has 128K of RAM (and slightly higher top speed as well along with more powerful core - Cortex M4)
edit - just looked on the code and there doesn't seem to be any part which would require more RAM as it operates on block size which is 2048B (2kB). Regardless there's evident drop of speed on BlueSCSI where there isn't one on SCSI2SD which indicates it handles bigger packets badly.
It's a wild guess but I imagine writing to an image file on the SD card has also implications on performance vs writing directly to the card as the image file is on a rather dumb file system that isn't aware of any wear leveling or any technique to use the free capacity to parallelize writes and I'm not sure if the rewriting of the blocks occours much more frequently than if you can use all available space on the disk. And I wonder if that leads to more fragmentation than using the whole SD.
@@gentuxable file system overhead shouldn't really affect performance that much and in such weird way. Wear leveling isn't filesystem feature but NAND controller and drive firmware. There's obviously some bottleneck when going from 512B to 32kB (and more). Wear leveling feature isn't also for gaining speed but extending lifetime of memory. In fact you need faster CPU core and more of them to make a decent NAND controller and it's not trivial to make it work on something like BluePill. But that aside - lack of wear leveling has nothing to do with performance issue here. And I don't think either filesystem or partition image file do. I'm curious where the fault lies but since I've got my hands full with my own business I will have to leave finding out to interface author.
BlueSCSI saved my marriage!
Have more BlueSCSIs than I'll likely ever need at this point because it was cheaper for me to build a bunch of them from parts (I think I ended up with about 20) than to buy a few outright.
Really the convenience of just being able to use an image file directly (or splicing out the partition when necessary) is a huge win for using these over most other solutions.
I had no idea that there were updates providing such a significant performance boost, so I'll definitely have to get on that...
Check the y axis on the three plots at 12:20 I think the spinny and scsi2sd did better than you think? or am I being daft?
I wouldn't consider PNY generic. Amazon has a bunch of truly sketchy no-brand SD cards that may or may not even have their listed storage capacity.
Great video, looking at blue scsi for some of my vintage computers
is that a Kitboga shirt?
WHY DID YOU PRESS REDEEM!? NOBODY SAID TO DO THAT!!! YOU DIDN'T HAVE TO DO THAT!!!
The t-shirt. 🤣
Great vid! Now I wish I could get Blue SCSI working like that with my SGI Indigo.
About 6:00 or so: TBW is Total Bytes Written, although usually it's displayed as TeraBytes Written, like on that card. Roughly, that little 128 GB SD card can handle 60 TB to be written to it in its lifetime, or rather, it's rated to endure that.
It is not likely to fail earlier than 60 TB. It may even last longer. But that's all they've certified it for.
The average 120/128GB SSD you use on your PowerPC Macs have a TBW rating of around ~200-300TB. More than the Macs will ever write. :)
A Hello from Brazil! I don't miss any of your videos.
Ola! Legal canal ne?
@@maxwellwagoner-watts4747 Muito! Me divirto demais!
Very cool.
Great Video!
Love the shirt!
Very nice. Now if you could compare the performance to RASCSI it would be even better.
I believe that the speed loss between the PowerPC and 68000 based processor is probably due to the fact that the PowerPC has to send disk data through a bridge and then onto a lower speed bus in order to reach the SCSI device. You don't see much difference with a mechanical hard disk as the bus speed is higher than the drive speed. But with super fast SD cards, the bus can run at it's maximum capacity so the speed is affected. Also, nowadays, even the cheapest SD cards probably outperform most old hardware. That's why the PNY card was not noticeably faster than the expensive ones.
The upgrade is in the CPU socket - same speed access to the logic board bus.
the PowerPC upgrade might be using a translation layer for the SCSI toolkit. That would have a big negative effect on speed.
@@phipli This reminds me a lot of the old 486 Pentium Overdrive which would greatly increase processing speed while choking on everything that was related to the external bus. But I was thinking about this. We don't really know how Mac Benchmark actually measures the speed. If it takes into account the bus speed on the accelerator card which gets choked-down on the 68000 bus, then the speed calculation drop will be directly measured and influence the final result.
@@MistahMatzah Apple never entirely removed legacy 68k code from the Mac OS codebase. They had mostly converted the OS to native PowerPC code by 8.6, but even the final 9.2.2 release still had some 68k code in it.
Apple had a built-in 68k emulator in Mac OS itself that targeted the 68LC040 CPU with some 68020/68030 features for additional backwards compatibility. Mac OS 8 and 8.1 was still almost entirely 68k code, so most everything was done in emulation.
@@Francois_L_7933 The only aspect of this machine that is choked down is the bus width. The PPC 601 has a 64 bit bus, while the 68LC040 has a 32 bit bus. Both CPUs have the same bus speed of 33 MHz, so you're looking at a 50% reduction in theoretical bandwidth. How much that aspect would otherwise affect performance is pretty negligible because no device on the bus is going to get anywhere close to saturating even the 68LC040's lower bus bandwidth.
Since the LC PDS slot is blocked by the upgrade board, the only two devices that will use any appreciable amount of bandwidth on the bus is the video controller and the SCSI controller. Since the video controller has its own VRAM, the only burden on the bus are draw calls and loading objects from the hard drive or system memory. The LC 575 still uses 8 bit single ended SCSI-1 with a maximum theoretical transfer rate of 5 MB/s.
The biggest offender here in limiting performance is Mac OS itself. Despite Apple transitioning to PowerPC in 1994, Mac OS had legacy 68k code in it all the way to the last 9.2.2 release. It wasn't until 8.6 that Apple eliminated the bulk of the legacy 68k code in Mac OS, meaning that the majority of the OS in older releases like 8.x on this mac heavily leaned on the integrated 68k emulator to work.
2:57 lol, I got the same box opener. It's a major improvement over the old metal one and using scissors
Is that Kitboga t-shirt? :D
So I says, "BlueSCSI, RedSCSI, they all wind up the same color in the end."
**eiffel 64 blue begins to play**
6:55 I don't quite follow, within a few seconds you say "the fastest card, which was the blue card", then "since the purple card did win, we'll continue using that one".
FYI older SD card speed ratings (U1, U2, U3) are optimized for sequential reads & writes in cameras, which is why the WD card is advertised "for smart video recording". More modern SD cards with an "A" rating will perform better on random reads & writes, particularly for things like apps and data on a smart phone.
Regardless, the old Mac and the BlueSCSI are not going to be fast enough to show the differences, but I'd probably use the Samsung card if I was handling disk images from another machine, like running images in a VM.
I would also go with Samsung. When it comes to SSDs or SD cards, the best brands in my opinion are the ones who make the memory controllers and chips themselves. So Samsung, Hynix, etc.
It's pretty much just the adapter itself and the card's lifespan TBW that matters in this case. Unless there's something wrong with the card itself, all new, name brand SD/uSD cards should outperform the I/O speed of the SCSI bus out of the box.
I just noticed your shirt. Cue the Kitboga vs Angry scammer montage. "WHY DID YOU REDEEM!?!?"
I am victorious with my 2015 MacBook Pro refurb!! Slipped a genuine Apple battery in there and it goes like a rocket now! Really quick... It has been a sweet project. The next piece of the project will be to replace the screen assembly but that can wait until it gets really bad, right now its just a couple of super fine vertical lines on the left and right hand side of the display. I plug it into my external monitor for serious work though so the built in screen is really just a place for my various social media apps to sit. Bought for $375 AUD, $148 AUD for a 1TB SSD, $20 for the SSD adapter. $87 AUD for a genuine apple battery. (It has a built in 8GB ram). Similar models are going for $1200 AUD at the moment. It's great for Logic Pro. ($675 total spend).
Could the speed difference between 68K and PPC be because the HD drivers are in 68K code that the PPC has to emulate? Or are the drivers fat binaries? (Or are the things Apple called 'drivers' not at all what we usually think of as drivers? That would be typical Apple...) I remember back in the day I ran little utilities on PPC Macs that flashed a little red indicator in the corner of the screen every time it was running 68K code, and I was appalled by how much legacy code was still seemingly running in MacOS 8 and even MacOS 9. I assume they were never really running 100% native code until OS X.
very interesting input.
HD drivers are more in the way of firmware -- they are not OS extensions.
The disk drivers are 68k code. If emulation overhead is the cause of the performance drop, then installing Connectix Speed Doubler should result in a big speedup. It could also just be because interrupt handling is a little more expensive on PowerPC.
A good test would be to update the disk drivers with FWB HDT 4.x, which is fully PPC native and optimized.
@@Stryder_The_Nite_Owl Will it still work when he switches back to 68k mode? That's the unique thing about this particular upgrade he's running.
Great Info Sean
Hi, great video. I am going to try this on the Mac IIsi I have coming. I was wondering what the best way to get unique or just old software off of the original scsi drives and archive or save them to be able to put it on the blue scsi?
Glad to know that your microSD extender isn't causing a difference. I do wonder if some are made cheaply tho, based on my prior tests on my full size SD model. (The ribbon is also way longer) I'll have to see if I can find it and do more testing!
They’re passive adaptors, right? There should be no speed penalty for adding some cable length. It’ll either work or it won’t.
Regarding the microSD card extender, I wonder if the wiring itself is shielded from EM? If not, maybe a simple aluminum foil wrap-around the cable might improve signal loss?
However, I agree. Those speed differences between the extender and not are very negligible. But...since you like to experiment, maybe this 5-minute project might be worth documenting?
OMG I love that shirt
Alright, I'm convinced. I have both a BlueSCSI and SCSI2SD in different machines. But for my SE/30 project underway now I'm going to go with the BlueSCSI. Real quick though... link for the extender cable? I like the idea of hanging mine out the back too.
pny sdcards are pretty common in asia. it's a brand enough to be a brand.
PNY is generally a common brand in industry and big setups
So, from the graphs, the Blue SCSI had a long way to go in order to match the speed of the SCSI2SD.
I so thought I was getting a Virtual PC video today, you faked me out, LOL!
Do you have disk cache in the memory control panel set on auto? As you switch between PPC and 68k and your RAM switches between 36 and 136 (or whatever it was), in auto, the disk cache would change and would impact performance.
Alternatively - I wonder is the driver you installed on the disk 68k code and running in emulation on the PPC?
Could you explore and show how the Asante SCSI Ethernet emulation works on there? I'm curious if it's worth getting old ethernet hardware for the classic mac or to just go with the Blue SCSI
Woah, is that finally working?
I really love the simplicity of the BlueSCSI. The speed not so much. SCSI2SD is still faster, especially v6. That said you could try the ZuluSCSI which combines the best of SCSI2SD with BlueSCSI simplicity. I have 2 of them but have not had a chance to test them yet. Speed was a big consideration when they created it so it should be fast then a V6. I hope so as we really need speed now.
"we're gonna get as much speed as possible out of this little blue device" sounds like something that'd get you put on a watchlist lol
@@youtubeisgarbage900 speed is another name for meth
@@youtubeisgarbage900 man it was a joke, no need to schizopost
@@youtubeisgarbage900 nah man, you just randomly brought up the "one world government" outta nowhere, I ain't the one wrong here
@@youtubeisgarbage900 what's your opinion on qanon?
@@youtubeisgarbage900 Somebody got blackpilled a little too hard. Lighten up, live a little, life sucks and the world is run by lunatics, doesn't mean you need to act like one too.
hahaha just caught the kitboga shirt thats awesome
I have found a Micro SD card compatible device which uses emmc storage. It is called the RaspiKey. It comes in 16 and 32 GB sizes. It is supposed to be faster and longer lasting than standard microSd cards. The key shape may be a problem for some situations, but your extender already handles that issue. Your info on the extenders not changing sdcard speed is very welcone I would rather beat up an extender than the built in slot in my machines. I may also get a full size to Micro sd card extender.
BTW the power card may be using the 68K for IO. Your speed loss may be handshaking between processors.
Any chance you could compare to a SCSI2SD v6?
more power - more limits.
TBW = Terrabyte Writes, so say if something has 80 TBW, thats how many terrabytes it can write over its life span before it really degrades if I recall correctly.
The biggest bottleneck with mechanical drives is not the read/write speed but the seek speed as it requires not only moving the head but additional rotations of the platter. This gets worse with the more platters you have in the drive as there is only one arm with two heads per platter so they all have to move furiously if the data block is spread across the platters, even if it has been defragmented. You can actually hear the amount of fragmentation on a drive by listening to the head swing back and forth.
With solid state media, the seek time is ZERO just like in RAM, which I don't think you benchmarked at all. The data is immediately accessible to the computer.
SCSI to sd for SGI systems needs more love. Lots of love for Mac and PC. But poor SGI...
i have a tough question for you, im stuck and maybe you know. Performa 6360 - is it possible to run 9.2.2 on it via os9 helper and a G3 L2 card? also, could this systems rom (version 2.0) be flashed to run another, say 2.1f that came on later PCI PPC macs so as to bypass the need for the os9 helper and allow it to run into os 10.4?
Scssi always sounded like a Muppet name to me lol
I don't have anything to say about the actual video topic, I just wanted to say nice Kitboga shirt.
EDIT: Oh hey and a classic seomoz shirt. I worked there for a year and a half in very recent history. Neat.
Hmmm, have you considered using on of those WiFi SD Cards on this? Basically you would be able to access the SD Images overr WiFi which can be quite an interesting feature, the thing runs Linux so you can get the correct drivers there to actually mount the images if need be etc ;)
Yo, nice Kitboga shirt!
is the native 68000 driver running in emulation on the PPC? just a thought
It is indeed emulated.
Well, I am delighted to say that while my comments from before weren’t the solution, I’m glad to have learned what the true solution was! …Now if I can just snag a BlueSCSI for myself!
it maske a massive diffrence when the card was made and what grade , none of them where graded for speed.
Still not sure why not just go up to a Sata ssd? One SCSI to IDE converter and a IDE to Sata converter would be enough lol. The latter I've used in several G3 and G4 machines, and they work like a dream. Very fast.
Does Kitboga sell t-shirts?
The current fastest SD-Card's are Amazon Basics - I dont know how reliable they are, it was a speed test, not a reliability test - The test was either based on or was, Jeff Geerlings SD-Card speed tests, and it was published about a month ago (mid - late March).
Yeah, you missed a crucial detail on the SCSI throughput test, the scales do not match between the left and right, one has 1024Kb as the Y axis maximum the other has a maximum of 2048Kb on the Y axis that one being (if I caught this right, the SCSISD and the BlueSCSI is the 1024Kb) This completely changes the results implications.
That is a noob mistake and actually also a trick tipically used in the news to make things to have the needed appearence to the audiences, not a nice trick and made worse when the scales ain't even present.
Not 100% sure but it looks like you were reading the y-axis of the charts incorrectly, as they have different scales
Purple SD Max Read ~870 Write ~660
SCSI2SD Max Read ~1340 Write ~1560
Where did you get the scammer t-shirt in the beginning of the video?
Man, I just opened the video and that shirt greeted me hahahahaha. Ayo, that is awesome, ain't that text from a TH-camr that was doing a granny voice and pissed some indian scammers? Hahahaha nice.
It is, yes. That TH-camr is Kitboga.
Micro sd cards, and probably readers, are super slow.. you should upgrade to SDXC and at least UHS-II, using Sandisk Pro cards, can give you upwards of 300 mb/second… I know, because I use them in my digital cameras, and the sandisk cards blow the other out of the water
You think that’s slow - you should see what SCSI devices in the early 90s peak at!
1MB/sec from a HDD in 1990-ish would be pretty good.
yeah... no... it won't make ANY difference. And yes, most of those cards presented in video were SDXC (except that PNY). Bottleneck isn't in card itself. Even SDHC achieves 2048+ (depending on class) KB/s sequential Read or Write (usually higher read than write). Bottleneck is elsewhere, not SD cards itself. Most likely both computer SCSI interface and SD-SCSI interface.
But is there a scsi2M.2NVME?
Maybe M.2 to sata to SCSI? I wonder what speed could be achieved
the t-shirt of yours is really funny :v
hey i think you forgot to add the info cards
OMG I had no idea that the SCSI/HDD drivers on classic Macs were literally on a hidden partition inserted by the disk initializing tool, with no (easy) way to replace or upgrade them without reformatting the drive from the scratch. It never ceases to amaze me just how much a pile of hacks Classic Mac OS was. Even Windows 9x seems a masterpiece of clean software architecture in comparison.
Knowing that, as for the 68k vs. PPC performance difference - maybe the disk drivers are 68k only? Or are forced into 68k mode due to the way the CPU switching works? My hypothesis is that the Toolbox ROM always boots in 68k mode due to the mainboard not understanding PPC natively, probably loads the SCSI drivers early in the boot process, and by the time later when the PPC CPU takes over, it's already too late to re-load the PPC version of the driver, so the 68k driver is emulated instead. I obviously don't know much about Classic Mac architecture, though, so this is just a wild guess.
lol dude your shirt! best scam bait ever.
Your shirt xD
MY PowerPC 9500/200 SCSI2SD V5.0a drive NO ???
NOOOO NOOOO NOOOOOOOOOO WHY DID YOU DO THAT MA'AM YOU DID NOT HAVE TO DO THAT YOU DID NOT HAVE TO REDEEM THE CARD
the cards he points to dont show up lol
wonder what would happen if you remove the sd ribbon wire and you went right to the board instead.
He did that in the video, it made no difference
I note early on, your T-shirt being a KitBoga reference. Nice to know ya support the scambaiter community! 😀
The PPC processor draws more power from the power supply thereby limiting other devices power inputs including the bluescsi.
I just posted a comment of what I think is going on with the slowdown you experienced on PPC, compared to 68k, at 7:23, but it's not showing up. I don't think you're immediately deleting or blocking my comments, but rather TH-cam is 'shadow-banning' me for directly including a hyperlink to a forum post, that i thought might explain what's happening.
In any case, there's a forum thread on the Avid Pro Audio Community website, titled "Nubus Power PC & Buffer Underrun", with a reply by someone named 'ulysses', from December 15, 1998. He describes a similar situation as you've shown in your benchmarking software. He was experiencing slow disk speeds in PPC, but when he disabled the PPC upgrade card, the disk speeds were much faster (he was burning images files, presumably from his hard disk to CD..... again, this is from 1998). To sum up what ulysses said, he essentially thinks that the PPC upgrade card he had installed (on his Quadra 950) had a flaw in the SCSI bus.
Yeah links get your comment automatically held for moderation now. And most TH-camrs never look in that tab of Studio, because it’s also where the spam and abusive comments get sent. IMO YT could do a better job of separating innocuous comments that have a link from the obvious spam…
Really?
You don’t know TBW??!!!
Terabytes Written!
(Life span)
Hi!
I'm not convinced about those SD card comparisons at all. Going from a class 4 to a class 10 card made a huge difference for me. Are you sure they're all class 10 cards? I suspect they're all 4.
The bottleneck here is not the card. Even the slowest is so much faster than the chain of hardware reading them that they’re all essentially equal.
This comment is community engagement. Have a nice day.
fourth