New user, Linux, 32-core Xeon working but not using GPUs

[kevvers]

I'm brand-new as of last night. I have a 32-core Xeon with GPUs available, but as near as I can tell it is not using the GPUs, of which there should be a lot.

The reason I think this is that the FAHControl SystemInfo windows down at the bottom says GPUs: 0 and OpenCL: Not detected: complaining libOpenCL.so cannot open shared object file because it does not exist.

I have been meaning to do a project with the GPUs but have been sidetracked. There's a large nVidia board in there that is just for the purpose. So some help with getting them up and running could make this much more productive.

[FldngForGrandparents]

Excuse Bruce, he replies to a lot of emails.

What version of Linux are you running? I assume Ubuntu? if so follow this and do the Graphics drivers and Open CL part:

https://linustechtips.com/main/topic/10 ... lding-rig/

[kevvers]

I'm running Linux (Xubuntu) so Windows update does not apply. I do run updates and seem to have current nVidia drivers: 440
I followed advice from @FidingForGrandparents and got new info in the log. It has changed, no longer complains about libioecl, but is otherwise inscrutable, so I'd appreciate a knowledgeable confirmation that my CUDA cores are in use
*********************** Log Started 2020-04-15T18:04:56Z ***********************
18:04:56:************************* Folding@home Client *************************
18:04:56: Website: https://foldingathome.org/
18:04:56: Copyright: (c) 2009-2018 foldingathome.org
18:04:56: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
18:04:56: Args: --child --lifeline 2414 /etc/fahclient/config.xml --run-as
18:04:56: fahclient --pid-file=/var/run/fahclient.pid --daemon
18:04:56: Config: /etc/fahclient/config.xml
18:04:56:******************************** Build ********************************
18:04:56: Version: 7.5.1
18:04:56: Date: May 11 2018
18:04:56: Time: 19:59:04
18:04:56: Repository: Git
18:04:56: Revision: 4705bf53c635f88b8fe85af7675557e15d491ff0
18:04:56: Branch: master
18:04:56: Compiler: GNU 6.3.0 20170516
18:04:56: Options: -std=gnu++98 -O3 -funroll-loops
18:04:56: Platform: linux2 4.14.0-3-amd64
18:04:56: Bits: 64
18:04:56: Mode: Release
18:04:56:******************************* System ********************************
18:04:56: CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
18:04:56: CPU ID: GenuineIntel Family 6 Model 79 Stepping 1
18:04:56: CPUs: 32
18:04:56: Memory: 251.79GiB
18:04:56: Free Memory: 250.25GiB
18:04:56: Threads: POSIX_THREADS
18:04:56: OS Version: 5.3
18:04:56: Has Battery: false
18:04:56: On Battery: false
18:04:56: UTC Offset: -7
18:04:56: PID: 2416
18:04:56: CWD: /var/lib/fahclient
18:04:56: OS: Linux 5.3.0-7648-generic x86_64
18:04:56: OS Arch: AMD64
18:04:56: GPUs: 0
18:04:56: CUDA Device 0: Platform:0 Device:0 Bus:3 Slot:0 Compute:5.2 Driver:10.2
18:04:56: CUDA Device 1: Platform:0 Device:1 Bus:2 Slot:0 Compute:3.5 Driver:10.2
18:04:56:OpenCL Device 0: Platform:0 Device:0 Bus:3 Slot:0 Compute:1.2 Driver:440.64
18:04:56:OpenCL Device 1: Platform:0 Device:1 Bus:2 Slot:0 Compute:1.2 Driver:440.64

[kevvers]

The thing that worries me most is that the log still includes (near the end):
18:04:56: GPUs: 0

[Joe_H]

The client configuration information is a bit further along in the log, if it includes a line of "gpu=false" a configuration change will need to be made before it will look for your GPU.

If that line is there, go to the Expert tab in Configure of FAHControl. Remove the option, and save the configuration change. Then restart FAHClient, or reboot. The client should then detect the GPU.

[kevvers]

Okay, I found

04:07:59: <!-- Folding Slot Configuration -->
04:07:59: <gpu v='false'/>

so I followed instructions. Not knowing how to stop and restart the client, I just did a system restart. Now I have
19:32:00: GPUs: 2
19:32:00: GPU 0: Bus:2 Slot:0 Func:0 NVIDIA:3 GK208 [GeForce GT 730]
19:32:00: GPU 1: Bus:3 Slot:0 Func:0 NVIDIA:6 GM200 [GeForce GTX 980 Ti] 5632
19:32:00: CUDA Device 0: Platform:0 Device:0 Bus:3 Slot:0 Compute:5.2 Driver:10.2
19:32:00: CUDA Device 1: Platform:0 Device:1 Bus:2 Slot:0 Compute:3.5 Driver:10.2
19:32:00:OpenCL Device 0: Platform:0 Device:0 Bus:3 Slot:0 Compute:1.2 Driver:440.64
19:32:00:OpenCL Device 1: Platform:0 Device:1 Bus:2 Slot:0 Compute:1.2 Driver:440.64
19:32:00:***********************************************************************

and

19:32:00:<config>
19:32:00: <!-- Client Control -->
19:32:00: <fold-anon v='true'/>
19:32:00:
19:32:00: <!-- Network -->
19:32:00: <proxy v=':8080'/>
19:32:00:
19:32:00: <!-- Slot Control -->
19:32:00: <power v='full'/>
19:32:00:
19:32:00: <!-- User Information -->
19:32:00: <user v='kevvers'/>
19:32:00:
19:32:00: <!-- Folding Slots -->
19:32:00: <slot id='0' type='CPU'>
19:32:00: <paused v='true'/>
19:32:00: </slot>
19:32:00:</config>

I'm not sure what this means, but I distrust type='CPU'

[kevvers]

And not to worry, I un-paused it.

[HaloJones]

If you're up to editing the config.xml, you could add the following lines immediately after the type=CPU:

Code: Select all

<slot id='0' type='GPU'>
  <opencl-index v='0'/>
</slot>
<slot id='1' type='GPU'>
 <opencl-index v='1'/>
</slot>

so pause the client, edit the config.xml, re-start the client.

[kevvers]

I'm up to it, but I cannot locate it. In /var/lib/fahclient/configure there are a number of xml files but I suspect they're historical. I don't see a plain config.xml.

Also I worry a bit about including both GPUs. One is connected to my monitor. The other is intended for precisely for this sort of thing.

[kevvers]

Based on @HaloJones' advice to pause, edit, re-start, I used FAHControl's configure, to add two -1 GPU slots.

WOW! Estimated daily points went from around 400 to around 3,000. That's what I had in mind!

TYVM.

[FldngForGrandparents]

You should expect to see a few hundred thousand with a GPU. The config file is located in /etc/fahclient

What does the command "nvidia-smi" show as the GPU name?

[anandhanju]

@kevvers, can you post your latest log? As FldngForGrandparents said, you should be getting a lot more than 3000 points with GPUs enabled for folding.

[kevvers]

Sorry, I thought I had posted an uodate that when that slot got a task it was over 100,000. All seems good

[Neil-B]

I think you may not be using a passkey … If this is the case, then without one (and other qualifications) you will simply receive base points for the project https://apps.foldingathome.org/psummary not the estimated ones shown in your interface … To understand more about the points system and how it works, including the Quick Return Bonus please see https://foldingathome.org/support/faq/points/

[HaloJones]

The 980TI is worth around 800K ppd (with a passkey) if it can get the work to keep it busy. Still a very capable card for folding.