Message boards : Number crunching : Confirm no nvidia GPU task
Message board moderation

To post messages, you must log in.

AuthorMessage
carmar

Send message
Joined: 3 Apr 22
Posts: 18
Credit: 1,774,000
RAC: 0
Message 5739 - Posted: 7 Aug 2022, 17:03:24 UTC

Sorry for the stupid question but I’d just like to confirm that only AMD GPU can process tasks but not NVIDIA, correct?

My conclusion is based off (a) the preferences parameters and (b) I’ve never seen my GPU being used (I have one machine with NVIDIA).

Thanks.
ID: 5739 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 10 May 20
Posts: 308
Credit: 4,733,484,700
RAC: 13,061
Message 5740 - Posted: 7 Aug 2022, 17:28:52 UTC - in response to Message 5739.  

Sorry for the stupid question but I’d just like to confirm that only AMD GPU can process tasks but not NVIDIA, correct?

My conclusion is based off (a) the preferences parameters and (b) I’ve never seen my GPU being used (I have one machine with NVIDIA).

Thanks.

Uhh, no. This is a cpu only project. No gpu applications here.

At other projects that DO use gpus, AMD, Intel and Nvidia gpus are used AS long as there are applications written for them.

The majority of projects use OpenCL applications which can be run by any gpu manufacturer.

Some projects provide CUDA applications which are specific to Nvidia. They are the fastest applications because there is no necessity for OpenCL translation.

If you aren't processing any gpu work on your Nvidia card, first thing to check is whether it is detected by BOINC. You can see that in the first 30 lines of the Event Log at BOINC startup.

If the project is using OpenCL apps, you need to see both CUDA and OpenCL functions be detected by BOINC at startup. If you only see CUDA function listed you need to install the OpenCL component of the video driver relative to your operating system.

It sounds like you are missing the OpenCL component for your drivers if you never process work on your Nvidia card yet are processing work for your AMD card. This assumes that the project application is OpenCL based.

A proud member of the OFA (Old Farts Association)
ID: 5740 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
carmar

Send message
Joined: 3 Apr 22
Posts: 18
Credit: 1,774,000
RAC: 0
Message 5741 - Posted: 7 Aug 2022, 17:57:24 UTC - in response to Message 5740.  
Last modified: 7 Aug 2022, 17:58:39 UTC

Thank you for your detailed explanation. Apologies again - my language was unclear.

I have one machine with Nvidia GPU and two with only integrated GPUs. I don’t have any machines with AMD GPUs. I have seen the nvidia GPU process tasks for other projects.

I did look at the event log but all I saw were lines about uploading tasks. However, you answered my question by confirming that only CPUs are used for this project, as well as clarified that not all applications for any project would use GPUs. Thanks again.
ID: 5741 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 10 May 20
Posts: 308
Credit: 4,733,484,700
RAC: 13,061
Message 5742 - Posted: 7 Aug 2022, 19:33:53 UTC - in response to Message 5741.  

You either need to scroll up to the beginning of the Event Log in the Manager to see the BOINC gpu detection report . . . or reboot the host to restart BOINC and then look at the beginning of the the Event Log. It will look something like this where the cards are first detected as CUDA capable and then reported as OpenCL capable.

03-Aug-2022 11:43:18 [---] Starting BOINC client version 7.19.0 for x86_64-pc-linux-gnu
03-Aug-2022 11:43:18 [---] This a development version of BOINC and may not function properly
03-Aug-2022 11:43:18 [---] log flags: file_xfer, sched_ops, task, sched_op_debug
03-Aug-2022 11:43:18 [---] Libraries: libcurl/7.68.0 OpenSSL/1.1.1f zlib/1.2.11 brotli/1.0.7 libidn2/2.2.0 libpsl/0.21.0 (+libidn2/2.2.0) libssh/0.9.3/openssl/zlib nghttp2/1.40.0 librtmp/2.3
03-Aug-2022 11:43:18 [---] Data directory: /home/keith/Desktop/BOINC
03-Aug-2022 11:43:18 [---] CUDA: NVIDIA GPU 0: NVIDIA GeForce RTX 2080 (driver version 510.85, CUDA version 11.6, compute capability 7.5, 7982MB, 7755MB available, 10598 GFLOPS peak)
03-Aug-2022 11:43:18 [---] CUDA: NVIDIA GPU 1: NVIDIA GeForce RTX 2080 (driver version 510.85, CUDA version 11.6, compute capability 7.5, 7979MB, 7567MB available, 10598 GFLOPS peak)
03-Aug-2022 11:43:18 [---] CUDA: NVIDIA GPU 2: NVIDIA GeForce RTX 2080 (driver version 510.85, CUDA version 11.6, compute capability 7.5, 7982MB, 7755MB available, 10598 GFLOPS peak)
03-Aug-2022 11:43:18 [---] OpenCL: NVIDIA GPU 0: NVIDIA GeForce RTX 2080 (driver version 510.85.02, device version OpenCL 3.0 CUDA, 7982MB, 7755MB available, 10598 GFLOPS peak)
03-Aug-2022 11:43:18 [---] OpenCL: NVIDIA GPU 1: NVIDIA GeForce RTX 2080 (driver version 510.85.02, device version OpenCL 3.0 CUDA, 7979MB, 7567MB available, 10598 GFLOPS peak)
03-Aug-2022 11:43:18 [---] OpenCL: NVIDIA GPU 2: NVIDIA GeForce RTX 2080 (driver version 510.85.02, device version OpenCL 3.0 CUDA, 7982MB, 7755MB available, 10598 GFLOPS peak)

You need to see BOTH Nvidia and OpenCL reports to be sure that the Nvidia card can run ANY gpu project. As mentioned previously in my first reply, the majority of gpu projects develop OpenCL applications as they are easier to write and are compatible across all gpu manufacturers. It takes strong Nvidia CUDA knowledge to properly develop CUDA scientific applications but the reward is much greater because of the 3X-10X performance gain over an OpenCL application using the same source code.

A common problem reported in the BOINC and project forums is the non-detection of Nvidia cards OpenCL capability and the report from people that they never receive any gpu work even though gpu work is allowed at a project. That is because the OpenCL component of the video drivers was never loaded. This is most often caused by Windows supplied video drivers which often exclude OpenCL capability. It is always best practice to load Nvidia drivers directly from the Nvidia download site and install those drivers over the ones that Microsoft supplies. Be wary of Microsoft also replacing the user supplied driver with their own without any notification to the user.

A proud member of the OFA (Old Farts Association)
ID: 5742 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
carmar

Send message
Joined: 3 Apr 22
Posts: 18
Credit: 1,774,000
RAC: 0
Message 5743 - Posted: 7 Aug 2022, 23:12:03 UTC - in response to Message 5742.  

Thank you for the walkthrough.

I checked again after a fresh boot. Both CUDA and OpenCL entries identify my nvidia GPU. I got one line for each. The event log refreshes very quickly and I couldn’t find the entries again but at least I can confirm.

Btw, I use a fork of Ubuntu, so no risk of MS drivers. But I appreciate the heads-up on directly getting drivers from nvidia site. When I used to be on Windows, I always used Nvidia site.
ID: 5743 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
3HkoYQNsAhAMofULHTjnCU8iHWrL

Send message
Joined: 19 May 22
Posts: 1
Credit: 422,667
RAC: 0
Message 5831 - Posted: 17 Sep 2022, 22:17:11 UTC - in response to Message 5739.  

sorry im have view right now your topic sorry for confusion im implement one more cpu on me network sytems sorry for inconvenient
ID: 5831 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Confirm no nvidia GPU task




Copyright © 2024 Copernicus Astronomical Centre of the Polish Academy of Sciences
Project server and website managed by Krzysztof 'krzyszp' Piszczek