Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Benchmarking 15 “E-Waste” GPUs with Modern Workloads

esologic.com

Post reply on HN
Print view View original

61–68 of 68 posts

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#61

Post by SillyUsername »

pjc50 wrote: post #55
joe_mamba wrote: post #39

Earlier quoted context omitted.

> about $80 (£60) Man, I wish I lived where you guys lived.

Yeah, that number doesn't align with the eBay listings I'm seeing. Perhaps the mere publication of this article has caused them all to sell out.

It was AliExpress, during sales, earlier this year, with discounts from playing games. I typically used to see much higher eBay prices. A bit of hyperbole tbh, not everyone has the patience to do this but I was honest that this is what I got them for. The DDR shortage has pushed the current price up to $130 (?) or >£85 after all discounts now.

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#62

Post by SillyUsername »

chromadon wrote: post #24
SillyUsername wrote: post #17

No mention of the venerable Tesla P4. 75W peak, 8GB VRAM, about $80 (£60). I have 6x P4s, a Xeon E5 2696v3 (36 threads, 3.8ghz peak but all core turbo unlocked, so 6 cores at 3.8Ghz - about 8 cores at 3.5ghz, or all cores at 3.1ghz), 48GB DDR4, all fit into a micro atx case running on a 650W MSI psu. This gives me a virtual 48GB GPU (llama.cpp ftw) to backup that 48GB of RAM. I typically see scores of at least 7-12t/…

How the hell did you fit 6 P4s in a mATX case?

Took off the bezels, compressed 4 into a block, had a rigid bifurcation card link GPUs 1 and 3 on top, another underneath the block do 2 and 4.

Then I fitted my own custom 40x40x30 fan shroud for 2x6000 rpm fans at one end.

The other 2 I did something similar for but a single 2 card block and a stock fan shroud online.

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#63

Post by SillyUsername »

chromadon wrote: post #24
SillyUsername wrote: post #17

No mention of the venerable Tesla P4. 75W peak, 8GB VRAM, about $80 (£60). I have 6x P4s, a Xeon E5 2696v3 (36 threads, 3.8ghz peak but all core turbo unlocked, so 6 cores at 3.8Ghz - about 8 cores at 3.5ghz, or all cores at 3.1ghz), 48GB DDR4, all fit into a micro atx case running on a 650W MSI psu. This gives me a virtual 48GB GPU (llama.cpp ftw) to backup that 48GB of RAM. I typically see scores of at least 7-12t/…

How the hell did you fit 6 P4s in a mATX case?

Oh and also side mounted a 4GB 750Ti for graphics, and/or whisper use, parallel to the mainboard. Airflow from the front fans to back is unimpeded by anything in this way.

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#64

Post by SillyUsername »

eso_logic wrote: post #19
SillyUsername wrote: post #17

No mention of the venerable Tesla P4. 75W peak, 8GB VRAM, about $80 (£60). I have 6x P4s, a Xeon E5 2696v3 (36 threads, 3.8ghz peak but all core turbo unlocked, so 6 cores at 3.8Ghz - about 8 cores at 3.5ghz, or all cores at 3.1ghz), 48GB DDR4, all fit into a micro atx case running on a 650W MSI psu. This gives me a virtual 48GB GPU (llama.cpp ftw) to backup that 48GB of RAM. I typically see scores of at least 7-12t/…

Yes! P4 and the other small cards are fantastic. I've been more focused on developing a cooling system around the 2 slot sized cards so I don't have any of these lying around. Are you using P4 for anything outside of LLM work? I'm interested in seeing their image processing capabilities.

Not really, I did consider some mining, just to see what the fuss was but probably wouldn't even break even.

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#65

Post by archi42 »

Azantys wrote: post #57
archi42 wrote: post #56

Earlier quoted context omitted.

Wait a few years, and we're all gaming on decomissioned data center GPUs ;-)

Old datacenter GPUs could game, but new ones can only do OpenCL/CUDA/ROCm etc. and have no display out. Im using an MI50 right now and I would wish newer datacenter cards could also be used for everything like them.

Ah, I'm aware of the physical limitations (including cooling), but I was under impression that these days we had a good handle on rendering on one device and doing the video output on another (at the expense of some latency). Obviously I never tried that with a datacenter GPU.

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#66

Post by Azantys »

archi42 wrote: post #65
Azantys wrote: post #57

Earlier quoted context omitted.

Old datacenter GPUs could game, but new ones can only do OpenCL/CUDA/ROCm etc. and have no display out. Im using an MI50 right now and I would wish newer datacenter cards could also be used for everything like them.

Ah, I'm aware of the physical limitations (including cooling), but I was under impression that these days we had a good handle on rendering on one device and doing the video output on another (at the expense of some latency). Obviously I never tried that with a datacenter GPU.

If you could render fully in OpenCL and output it via another device it would probably work. But they dont have ROPS so they cant do traditional video rendering. Even something like Blender where you would use it for calculations and not for display/video doesnt work without software emulating hardware features related to textures.

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#67

Post by kasabali »

latchkey wrote: post #21
riedel wrote: post #20

Earlier quoted context omitted.

Cool stuff. Just read: https://github.com/akandr/bc250

Yea, I'm bummed I didn't know about the 40-CU unlock, although it probably wouldn't have had much impact on mining performance. It still would have been neat to test. I did build a whole automated solution for auto-tuning each individual board. It would start at the "best" settings and then downgrade every time there was a crash. If it wasn't crashing, then those were the new "best" settings for that individual chip.

afaik unlock was discovered very recently

Re: Benchmarking 15 “E-Waste” GPUs with Modern Workloads

#68

Post by latchkey »

kasabali wrote: post #67
latchkey wrote: post #21

Earlier quoted context omitted.

Yea, I'm bummed I didn't know about the 40-CU unlock, although it probably wouldn't have had much impact on mining performance. It still would have been neat to test. I did build a whole automated solution for auto-tuning each individual board. It would start at the "best" settings and then downgrade every time there was a crash. If it wasn't crashing, then those were the new "best" settings for that individual chip.

afaik unlock was discovered very recently

It was discovered recently, however we were dealing with asrock and AMD directly given the quantity we bought, and they didn’t say anything. Kind of lame imho.
Post reply on HN