Page 1 of 1

numbers in GPUs.txt

PostPosted: Mon Aug 28, 2017 3:44 pm
by ChristianVirtual
Saw recently number appearing after the GPU name ... what is the purpose for those ?

0x10de:0x17c2:2:7:GM200 [GeForce GTX Titan X] 6144
0x10de:0x17c8:2:7:GM200 [GeForce GTX 980 Ti] 5632
0x10de:0x17f0:2:5:GM200GL [Quadro M6000]
0x10de:0x17f1:2:5:GM200GL [Quadro M6000]
0x10de:0x17fd:2:5:GM200GL [Tesla M40]
0x10de:0x1b00:2:7:GP102 [GeForce Titan X] 6144
0x10de:0x1b01:2:5:GP102 [xxx]
0x10de:0x1b02:2:7:GP102 [GeForce Titan Xp] 12150
0x10de:0x1b06:2:7:GP102 [GeForce GTX 1080 Ti] 11380
0x10de:0x1b30:2:5:GP102GL [Quadro P6000]
0x10de:0x1b38:2:5:GP102GL [Tesla P40]

Re: numbers in GPUs.txt

PostPosted: Mon Aug 28, 2017 6:51 pm
by bruce
They're the expected SP-GFLOPS. They're not used by FAH but they do help us sort out potential errors in GPUSpecies.

Re: numbers in GPUs.txt

PostPosted: Mon Aug 28, 2017 9:36 pm
by ChristianVirtual
Thanks Bruce, at least I know they are on purpose and can tweak my iPad app to ignore them (not messing my internal DB)

Re: numbers in GPUs.txt

PostPosted: Mon Aug 28, 2017 9:49 pm
by Joe_H
Everything to the right of the last colon is "cosmetic", does not actually get used by the folding client at all for other than for display purposes. So I would not place any particular importance on that field in your app.

Re: numbers in GPUs.txt

PostPosted: Mon Aug 28, 2017 11:18 pm
by ChristianVirtual
Beside that's the only place from the log (3rd part API or file upload) where I get the name from ...

http://imageshack.com/a/img923/5876/LZ08hF.jpg

But hey, no difficult to cut everything after the squared bracket

Re: numbers in GPUs.txt

PostPosted: Sat Feb 15, 2020 6:09 pm
by ari2asem
is there support for nvidia grid m40?

not asking for support tesla m40. but for grid m40....

Re: numbers in GPUs.txt

PostPosted: Sat Feb 15, 2020 6:52 pm
by bruce
The GM107GL [GRID M40] is listed as Species 4, which means it's recognized as a GPU that doesn't support Double Precision so it won't get any assignments. I have not found a clear description of what it can/cannot do.

Edit: Found it. The GRID M40 contains a GM200 GPU and can do 7 TFLOPS Single Precision (FP32) and 0.21 TFLOPS (1/32) Double Precision (FP64). That's not what GPUS.TXt current says, but I can correct it if you have one that can be tested. It was designed for Machine Learning, not for stream processing, and I expect it will perform VERY poorly on FAH but we won't know that unless you can report back.

Re: numbers in GPUs.txt

PostPosted: Sun Feb 16, 2020 1:37 pm
by toTOW
The one I found is way less powerful : https://www.techpowerup.com/gpu-specs/grid-m40.c2518 ... it's a GM107, not a GM200 ...

FP32 (float) performance
793.3 GFLOPS

FP64 (double) performance
24.79 GFLOPS (1:32)


It should fold, but don't expect high performance.