Advanced search

Message boards : Graphics cards (GPUs) : Any ideas why I'm not getting GPU tasks?

Author Message
widdershins
Send message
Joined: 30 Jun 10
Posts: 5
Credit: 18,115,721
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwat
Message 55062 - Posted: 21 Jun 2020 | 0:18:48 UTC

Ok, some background. Joined GPU grid many years ago, but graphics card was hopeless. I can't remember now whether GPU grid stopped sending me units because my card no longer met minimum requirements, or if I stopped because units were taking too long to complete. Been happily crunching on WCG with cpu however.

So on to today, recently replaced an old computer with a new(er) one via ebay and got a new(er) titan x as well.

Have been running WCG COVID units just fine and thought I'd try my hand at GPU Grid again. Added to boinc manager projects, logged in to GPUGrid control panel and selected all projects, ticked the gpu box, then sat back and waited, and waited. Nothing. Servers seem to be all green, units seem to be waiting to go out, but I'm getting nothing. As I see it the main possibilities are units that are available are not for linux, or my newer GPU is still too underpowered.

In case it helps with any suggestions regarding what I'm doing wrong here's some info on the O/S and GPU

OS info:

mint Kernel: 5.0.0-32-generic x86_64 bits: 64 compiler: gcc v: 7.4.0
Desktop: Cinnamon 4.4.5 wm: muffin dm: LightDM
Distro: Linux Mint 19.3 Tricia
base: Ubuntu 18.04 bionic

GPU info:

Device-1: NVIDIA GM200 [GeForce GTX TITAN X] driver: nouveau v: kernel bus ID: 05:00.0
chip ID: 10de:17c2
Display: x11 server: X.Org 1.20.4 driver: modesetting unloaded: fbdev,vesa
Driver: Nvidia 435.21-0ubuntu0.18.04.2
resolution: 1920x1080~60Hz
OpenGL: renderer: NV120 v: 4.3 Mesa 19.2.8 direct render: Yes

All suggestions welcome - well almost all ;-)

Thanks

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 808
Credit: 1,076,309,831
RAC: 924,854
Level
Met
Scientific publications
watwatwatwatwat
Message 55063 - Posted: 21 Jun 2020 | 0:46:23 UTC - in response to Message 55062.

First read this post on the FAQ forum about the requirements to run the acemd3 application.

https://www.gpugrid.net/forum_thread.php?id=5002#52865

From your post you are not running any Nvidia proprietary drivers. You are running the stock Nouveau drivers which CANNOT do compute as they have no CUDA support or OpenCL support. They can only drive an image to a monitor.

You need to blacklist the Nouveau drivers so that only the proprietary Nvidia drivers are loaded.

I see that you have the 435.21 drivers loaded but are not being used.

Also you cannot get any gpu work for any project until BOINC has detected a suitable gpu card at startup. You should see a printout in the BOINC Event Log stating that both CUDA and OpenCL drivers have been detected. This project requires CUDA drivers, but most of the other gpu projects require OpenCL drivers for their applications. This is an example of what the beginning of the Event Log would show for properly detected gpus.

20-Jun-2020 16:49:56 [---] Data directory: /home/keith/Desktop/BOINC
20-Jun-2020 16:49:57 [---] CUDA: NVIDIA GPU 0: GeForce RTX 2080 (driver version 440.64, CUDA version 10.2, compute capability 7.5, 7982MB, 7743MB available, 10598 GFLOPS peak)
20-Jun-2020 16:49:57 [---] CUDA: NVIDIA GPU 1: GeForce RTX 2080 (driver version 440.64, CUDA version 10.2, compute capability 7.5, 7979MB, 7537MB available, 10598 GFLOPS peak)
20-Jun-2020 16:49:57 [---] CUDA: NVIDIA GPU 2: GeForce RTX 2080 (driver version 440.64, CUDA version 10.2, compute capability 7.5, 7982MB, 7743MB available, 10598 GFLOPS peak)
20-Jun-2020 16:49:57 [---] OpenCL: NVIDIA GPU 0: GeForce RTX 2080 (driver version 440.64, device version OpenCL 1.2 CUDA, 7982MB, 7743MB available, 10598 GFLOPS peak)
20-Jun-2020 16:49:57 [---] OpenCL: NVIDIA GPU 1: GeForce RTX 2080 (driver version 440.64, device version OpenCL 1.2 CUDA, 7979MB, 7537MB available, 10598 GFLOPS peak)
20-Jun-2020 16:49:57 [---] OpenCL: NVIDIA GPU 2: GeForce RTX 2080 (driver version 440.64, device version OpenCL 1.2 CUDA, 7982MB, 7743MB available, 10598 GFLOPS peak)


Once you get BOINC to detect your Nvidia gpu, the project will send you work. There is plenty currently.

Good luck.

Aurum
Avatar
Send message
Joined: 12 Jul 17
Posts: 307
Credit: 10,279,318,231
RAC: 0
Level
Trp
Scientific publications
watwatwat
Message 55064 - Posted: 21 Jun 2020 | 2:33:47 UTC - in response to Message 55063.

You need to blacklist the Nouveau drivers so that only the proprietary Nvidia drivers are loaded.
No do NOT try to blacklist nouveau in Linux Mint 19.3. Just open Driver Manager and click 440.82 the latest Nvidia driver, install & reboot.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 808
Credit: 1,076,309,831
RAC: 924,854
Level
Met
Scientific publications
watwatwatwatwat
Message 55065 - Posted: 21 Jun 2020 | 16:27:04 UTC - in response to Message 55064.

You need to blacklist the Nouveau drivers so that only the proprietary Nvidia drivers are loaded.
No do NOT try to blacklist nouveau in Linux Mint 19.3. Just open Driver Manager and click 440.82 the latest Nvidia driver, install & reboot.

He already has sufficient Nvidia drivers loaded. He says the 435.21 drivers are installed.

Yet they are not being used. He is using the Nouveau drivers and that is insufficient for BOINC or the project to use his card.

If he blacklisted the Nouveau driver, it would fall back on the proprietary Nvidia driver and all would be well.

The first thing that installing the Nvidia drivers does is blacklist the Nouveau driver.

So yes, if he installed the latest 440.82 drivers, that would get rid of the Nouveau drivers and the driver situation would be sorted out.

Either way would work.

Aurum
Avatar
Send message
Joined: 12 Jul 17
Posts: 307
Credit: 10,279,318,231
RAC: 0
Level
Trp
Scientific publications
watwatwat
Message 55066 - Posted: 22 Jun 2020 | 16:32:40 UTC
Last modified: 22 Jun 2020 | 16:41:32 UTC

Linux Mint eliminated any need to do the blacklisting kluge a long time ago. It's as simple as clicking the radio button of the desired Nvidia driver from the repository. I assume 435 works but I keep current myself.


If my screenshot doesn't display it's here:
https://i.ibb.co/3yWr09m/Nvidia-Driver-Manager.jpg

Ian&Steve C.
Avatar
Send message
Joined: 21 Feb 20
Posts: 539
Credit: 4,449,206,357
RAC: 679,821
Level
Arg
Scientific publications
wat
Message 55067 - Posted: 22 Jun 2020 | 17:19:41 UTC - in response to Message 55066.

Linux Mint eliminated any need to do the blacklisting kluge a long time ago. It's as simple as clicking the radio button of the desired Nvidia driver from the repository. I assume 435 works but I keep current myself.


If my screenshot doesn't display it's here:
https://i.ibb.co/3yWr09m/Nvidia-Driver-Manager.jpg


it's the same with Ubuntu 18 and 20 really. you can click the little easy-buttons if you want to.

it comes down to how the driver was installed. if you use a package manager to install it, you usually don't have to manually blacklist the nouveau driver. but if you installed from the nvidia provided run file for linux (needed if your distribution doesn't have a package available in the repos or PPAs, or if you want a beta or newer driver version not provided by your available repos) then you likely will have to manually blacklist, even on Linux Mint.

____________

widdershins
Send message
Joined: 30 Jun 10
Posts: 5
Credit: 18,115,721
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwat
Message 55068 - Posted: 22 Jun 2020 | 23:36:51 UTC

After reading the comments and some further attempts and some googling I think I understand the problem and may have missed a piece of information which would have the more experienced linuxers on here laughing.

I've been running it via a live CD since one of the projects I've been running on WCG is fond of disk writes and with 24 units going simultaneously I figured I'd save my disk a lot of wear since I have loads of ram the live distro can use. Been using it for years without problems. So when I decided to try GPUgrid again I just added the Boinc cuda package to my current session, that the nvidia driver, checked driver manager and it said the Nvidia driver was installed and in use.

After trying reinstalling the newer drivers and then googling following the comments here that the driver wasn't being used I now realise that driver manager lies when in live mode. From what I've read it's difficult if not impossible to get the cuda driver running in live mode unless you have a black belt in linux - which I don't sadly, I'm typing one handed as I'm holding my trousers up with the other! :-D

So I'm going to let the WCG units run out, and switch back to windows, alter my mix of WCG units and retry GPUgrid. Hopefully I'll have more success.

Thanks for all your comments they've been very helpful in identifying where I was going wrong. I was afraid it was my own stupidity that was to blame, but was hoping it wasn't - for once. :-(

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 808
Credit: 1,076,309,831
RAC: 924,854
Level
Met
Scientific publications
watwatwatwatwat
Message 55069 - Posted: 23 Jun 2020 | 1:19:43 UTC - in response to Message 55068.

Yes, a critical piece of information omitted. I don't think any of us expected someone to be running off a LiveCD.

Lots of stuff won't run correctly as the applications expect a physical hard drive for storage to use the LiveCD in Persistence mode.

The LiveCD is mainly just for advertisement and test tryouts. Oh, and emergency drive maintenance.

I don't think you realize how robust modern storage is for drive writes. Even flash storage is rated in TBW or terabytes written or full drive writes per day.

widdershins
Send message
Joined: 30 Jun 10
Posts: 5
Credit: 18,115,721
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwat
Message 55087 - Posted: 28 Jun 2020 | 11:35:17 UTC

Me again, thanks to you folks I seem to have GPU grid crunching GPU units, on Win 10 rather than the live linux CD. A bit of messing about was required though.

First after downloading the CUDA package from Nvidia I tried just installing the graphics driver since I figured that would be all that was required as I wasn't developing new cuda software. Nope! So re-installed with everything but the Nvidia Geforce experience and that worked, to a degree. Now boinc said cuda was there but it couldn't use it. Cue more googling, and after letting WCG WU run out again I uninstalled BOINC and reinstalled in user mode rather than leaving the little tick box pre-checked.

Rebooted everything and success!!!

It might be worth creating an idiots guide to getting GPU Grid up and running on different operating systems and sticking it on the website. There are a lot more of us out there than you'd think. :-D

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 808
Credit: 1,076,309,831
RAC: 924,854
Level
Met
Scientific publications
watwatwatwatwat
Message 55088 - Posted: 28 Jun 2020 | 12:23:39 UTC

It is always hard to help someone who has hidden their computers because we have no idea what hardware or software they are running.

Normally it is not very hard at all to get running Nvidia drivers on Linux because they are installed by default by the OS usually or at least are an easy point and click install from the distro.

The absolute hardest method to install Nvidia drivers is to download and install the drivers direct from Nvidia as their installation method is hard and hard to understand and follow for beginners. It sounds like that was your case. It would normally have not been necessary unless you are using a very old or different distro compared to recent popular distros.

Post to thread

Message boards : Graphics cards (GPUs) : Any ideas why I'm not getting GPU tasks?