Old Quadro 4000 (GF100): Should I expect only 9405PPD?

[starkruzr]

Hi,
Have an old HP workstation I've put into service folding. The CPU (an E5 Ivy Bridge Xeon) gets about 41605PPD, but the GPU, so far, reports 9405PPD. I know this is a super old card, but can that be right? I would think more or less any supported GPU would be way faster than this.
Thanks.

[PantherX]

Welcome to the F@H Forum starkruzr,

Please note that Quadro 4000 doesn't support OpenCL 1.2 which is a requirement now for GPUs: https://www.techpowerup.com/gpu-specs/quadro-4000.c898

You can use all your CPUs for folding and ensure that you have a passkey if you would like to get bonus points: https://foldingathome.org/support/faq/points/passkey/

[starkruzr]

Welcome to the F@H Forum starkruzr,

Please note that Quadro 4000 doesn't support OpenCL 1.2 which is a requirement now for GPUs: https://www.techpowerup.com/gpu-specs/quadro-4000.c898

You can use all your CPUs for folding and ensure that you have a passkey if you would like to get bonus points: https://foldingathome.org/support/faq/points/passkey/

Thanks, I've done that already. I guess it's more strange that it works at all than that I'm only getting ~10K PPD.

[Joe_H]

as best as I can tell from the specs, it is about the equivalent of a GT 460 card, so that sounds about in the right range. It is possible the PPD might show a bit higher, depends on the WU and how long it had been running.

[starkruzr]

So, interesting update here: today, it says it's doing 31K PPD. Not bad! Is there some arrangement where WUs are specific to libraries -- like, maybe this one is CUDA-specific and that works fine on this old Quadro, so the OpenCL1.2 limitation doesn't matter?

[PantherX]

WUs vary depending on the number of atoms they have. Smaller atoms are good for low-end GPUs, not so much for high-end GPUs. High atom counts are great for high-end GPUs but not so good for low-end GPUs. You can find the atom number by looking up the Project number here: https://apps.foldingathome.org/psummary

[starkruzr]

WUs vary depending on the number of atoms they have. Smaller atoms are good for low-end GPUs, not so much for high-end GPUs. High atom counts are great for high-end GPUs but not so good for low-end GPUs. You can find the atom number by looking up the Project number here: https://apps.foldingathome.org/psummary

Thanks, that's great to know!

[RichieOnTheFold]

Hi there,

I have the same graphics card and was successfully folding with the GPU until a few days ago. I had upgraded the FAH client from 7.6.9 to 7.6.13 and I think the GPU stopped folding soon after that. I've tried everything I can think of but I can't get it folding again. Can you share the version number of the nVidia driver you are using and any other tips to get the thing folding again? Cheers.

[JimboPalmer]

RichieOnTheFold, one thing that sometimes happens is that Microsoft runs Windows Update and updates your driver, always download new drivers from Nvidia.

If you have 64 bit Windows 10, this might be the right link: https://www.nvidia.com/Download/driverR ... 0177/en-us

I am suprised it folds at all as it does not support OpenCL 1.2 which is the minimum requirement, and suprised again that it completes on time, it is about 3 times as slow as a GTX 1030, the slowest modern graphics card Nvidia will sell you.

[RichieOnTheFold]

Thanks, that is the driver that I re-installed.

It is my understanding that the Quadro 4000 is supported, and the fact that it was folding on the GPU until just recently makes me think that something else changed but I can't figure out what. Since the GPU is significantly faster than the CPU, I wouldn't think that they'd disable usage of the GPU yet still fold on the CPU, but who knows. Any other thoughts? I can post any diagnostics that might help. Cheers.

[JimboPalmer]

Since the GPU is significantly faster than the CPU, I wouldn't think that they'd disable usage of the GPU yet still fold on the CPU, but who knows. Any other thoughts? I can post any diagnostics that might help. Cheers.

CPUs have not changed much in the last two decades Core_a7 supports CPUs back to 2001. (Pentium 4 Willamette) If the CPU supports AVX (starting in 2011) Core_a7 is faster yet, but it supports CPUs from this century. Intel helped F@H write code for CPUs stating in 2001.

GPUs are changing and becoming more sophisticated much more rapidly. (I am going to use Nvidia examples, but AMD is much the same)

In 2007, Nvidia made its first Programmable Graphics card. The GT 8000 series

2007 Nvidia began to assist F@H to write code for GPUs.

Currently, F@H requires both OpenCL 1.2 and Double Precision floating point math. (FP64)

The earliest GPUs that meet both requirements were introduced in March 2010. (GTX 465,GTX 470, GTX 480)
The last Desktop GPU that did not support Double Precision may have been the GT 645 in April 2012
The last Desktop GPU that did NOT support OpenCL 1.2 may have been the GT 730 in June 2014
So cards designed after June 2014 should fold, however Laptop chips will lag. (2017?)
Workstation cards are not my price range, but should be similar to Desktop cards.

https://en.wikipedia.org/wiki/List_of_N ... sing_units
https://foldingathome.org/about/partners/

[RichieOnTheFold]

Hi again,

The GPUs.txt file in the FAH Data Directory contains this line:
0x10de:0x06dd:2:2:GF100 [Quadro 4000]

I don't have a key to decoding that line, but somewhere on this forum someone said that if the position where the 2s are contains blanks then that GPU is not supported by FAH. So I presume that since the Quadro 4000 has 2s in those positions, then FAH supports that GPU. Besides, the proof is in the pudding; the GPU was folding up until just a few days ago.

I've already tried uninstalling & reinstalling both the FAH Client and the nVidia drivers on my workstation. I also tried reverting back to v7.6.9 from 7.6.13. I also tried various older nVidia drivers and the Win10 driver (updated driver from within Device Manager). I checked the firewall and it looks fine (same as my laptop which continues to fold with both CPU and GPU). The CPU folds on my workstation, but the GPU doesn't. Here are a few lines from the log file:
01:48:51:WARNING:WU00:FS01:WorkServer connection failed on port 8080 trying 80
01:48:51:WU00:FS01:Connecting to 192.0.2.1:80
01:49:12:ERROR:WU00:FS01:Exception: Failed to connect to 192.0.2.1:80: A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

Does anyone know how to troubleshoot this further?

Cheers

[JimboPalmer]

I am not very savvy about those positions but i am going to give you my SWAG.

0:0: or the equivalent :: may mean it has never been able to fold. 192.0.2.1 is a code for unsupported.

I think species 4, is no longer supported, and species 5 is supported. I am unclear which of the two is the species number. I know that the new AMD Navi cards (RDNA) has to get a new species as Core_21 is not supported by them, so Species 7 can use Core_22 but not use Core_21. I do not think Nvidia has made changes since 2014 (see above) that obsolete any cards. (Some cards that meet all the requirements may still not complete in time to Fold)

There is a new OpenCL 3 standard coming that may obsolete some older graphics cards but the standard almost says it can work with OpenCL 1.2 devices. We shall see. F@H holds on to 'obsolete' equipment as long as it can. and then there is a backlash when they finally give it up.

(Not hardware, but it is getting harder and harder to fold on a 32 bit OS. GPUs are not supported well, the drivers are hard to find. Again, this is not F@H's doing. Time marches on.)

[RichieOnTheFold]

Well, thanks anyway. I'll just uninstall FAH from my workstation and keep it running on my laptop. Cheers.

[JimboPalmer]

It should still work for CPUs.