Page 1 of 2
1080 Nvidia GPU not available - Fedora 31
Posted: Mon Mar 09, 2020 10:20 pm
by GEI
Installed client & control rpms, opencl & cuda are both installed along with the newest Nvidia drivers for Fedora 31
FAH continues to log "GPU not available" error.
Any ideas on what to track down to get GPU working with FAH?
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Mon Mar 09, 2020 11:12 pm
by bollix47
Welcome to the folding@home community driven support forum.
First, did you reboot after installing the drivers?
If yes then open a terminal and type the following then post the output here:
FAHClient --lspci
You could also show us your log:
viewtopic.php?f=24&t=26036
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 12:02 am
by GEI
Log:
Code: Select all
*********************** Log Started 2020-03-09T23:56:23Z ***********************
23:56:23:************************* Folding@home Client *************************
23:56:23: Website: https://foldingathome.org/
23:56:23: Copyright: (c) 2009-2018 foldingathome.org
23:56:23: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
23:56:23: Args: --child --lifeline 78225 /etc/fahclient/config.xml --run-as
23:56:23: fahclient --pid-file=/var/run/fahclient.pid --daemon
23:56:23: Config: /etc/fahclient/config.xml
23:56:23:******************************** Build ********************************
23:56:23: Version: 7.5.1
23:56:23: Date: May 12 2018
23:56:23: Time: 22:51:07
23:56:23: Repository: Git
23:56:23: Revision: 4705bf53c635f88b8fe85af7675557e15d491ff0
23:56:23: Branch: master
23:56:23: Compiler: GNU 4.4.7 20120313 (Red Hat 4.4.7-18)
23:56:23: Options: -std=gnu++98 -O3 -funroll-loops
23:56:23: Platform: linux2 4.14.0-3-amd64
23:56:23: Bits: 64
23:56:23: Mode: Release
23:56:23:******************************* System ********************************
23:56:23: CPU: Intel(R) Xeon(R) CPU E5-2687W 0 @ 3.10GHz
23:56:23: CPU ID: GenuineIntel Family 6 Model 45 Stepping 7
23:56:23: CPUs: 32
23:56:23: Memory: 62.83GiB
23:56:23: Free Memory: 41.79GiB
23:56:23: Threads: POSIX_THREADS
23:56:23: OS Version: 5.5
23:56:23: Has Battery: false
23:56:23: On Battery: false
23:56:23: UTC Offset: -7
23:56:23: PID: 78227
23:56:23: CWD: /var/lib/fahclient
23:56:23: OS: Linux 5.5.5-200.fc31.x86_64 x86_64
23:56:23: OS Arch: AMD64
23:56:23: GPUs: 0
23:56:23:CUDA Device 0: Platform:0 Device:0 Bus:5 Slot:0 Compute:6.1 Driver:10.2
23:56:23: OpenCL: Not detected: clGetDeviceIDs() returned -1
23:56:23:***********************************************************************
23:56:23:<config>
23:56:23: <!-- Folding Slot Configuration -->
23:56:23: <gpu v='false'/>
23:56:23:
23:56:23: <!-- Network -->
23:56:23: <proxy v=':8080'/>
23:56:23:
23:56:23: <!-- User Information -->
23:56:23: <user v='anonymous'/>
23:56:23:
23:56:23: <!-- Folding Slots -->
23:56:23: <slot id='0' type='CPU'>
23:56:23: <cpus v='8'/>
23:56:23: </slot>
23:56:23:</config>
23:56:23:Switching to user fahclient
23:56:23:Trying to access database...
23:56:23:Successfully acquired database lock
23:56:23:Enabled folding slot 00: READY cpu:8
23:56:23:WU00:FS00:Starting
23:56:23:WU00:FS00:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/cores.foldingathome.org/v7/lin/64bit/avx/Core_a7.fah/FahCore_a7 -dir 00 -suffix 01 -version 705 -lifeline 78227 -checkpoint 15 -np 8
23:56:23:WU00:FS00:Started FahCore on PID 78286
23:56:23:WU00:FS00:Core PID:78290
23:56:23:WU00:FS00:FahCore 0xa7 started
23:56:24:WU00:FS00:0xa7:*********************** Log Started 2020-03-09T23:56:23Z ***********************
23:56:24:WU00:FS00:0xa7:************************** Gromacs Folding@home Core ***************************
--lspci
..................
0x8086:0x1d08:0:31:5:Intel Corporation:
0x1000:0x0086:2:0:0:LSI Logic / Symbios Logic:
0x10de:0x1b80:5:0:0:NVIDIA Corporation:
0x10de:0x10f0:5:0:1:NVIDIA Corporation:
0x8086:0x1d6b:4:0:0:Intel Corporation:
0x8086:0x10d3:1:0:0:Intel Corporation:
0x104c:0x8241:8:0:0:Texas Instruments:
0x11c1:0x5811:9:1:0:LSI Corporation:
0x8086:0x3c04:128:2:0:Intel Corporation:
.................................
rest are Intel cores
rebooted, checked all drivers
I need the GPU to make it usable, otherwise if strictly CPU the core temperatures soar into upper 80c range.
Mod Edit: Added Code Tags - PantherX
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 12:28 am
by bollix47
See if you can install a package called ocl-icd-opencl-dev ... that's what it's called in debian but I don't have fedora knowledge so you may have to 'dig' a bit.
Once you've got it installed reboot and look at the log to see if the following changes:
23:56:23: GPUs: 0
23:56:23:CUDA Device 0: Platform:0 Device:0 Bus:5 Slot:0 Compute:6.1 Driver:10.2
23:56:23: OpenCL: Not detected: clGetDeviceIDs() returned -1
Also, where did you get the GPU drivers? ie normal repo, graphic drivers repo or nvidia.com/drivers ?
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 1:23 am
by GEI
Had mesa-OpenCL installed, installed meas-OpenCL-devel
rebooted
GPU now shows, though OpenCL throughs error and is still -1
I can now "build" a gpu slot, but it just seems to sit there in ready state without picking up the job. Does each slot run its own job?
NVidia drivers are repackaged repo I have been using for years with Tensorflow-gpu successfully.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 1:52 am
by Joe_H
Since the client did not detect a GPU when installed, the configuration has saved that as shown by:
Code: Select all
23:56:23:<config>
23:56:23: <!-- Folding Slot Configuration -->
23:56:23: <gpu v='false'/>
23:56:23:
You can reset that by going to the Expert tab of Configure in FAGControl and entering in the option "gpu" set to the value of "true" as Extra Client Option. Save, and restart the client. You should then be able to set up a GPU slot and get assignments.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 2:03 am
by GEI
I had tried this approach before originally posting. Couldn't find the GPU and reset gpu v=false
After a number of restarts and the mesa-OpenCL-devel install the GPU now has a separate job and is working on it.
Running under medium yields 195507 pts/day while keeping CPU temps under 80c
Thanks for the inputs. I'll leave it set as is for now and let it run for 24hrs and see how it looks.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 2:39 am
by bruce
Are you sure you installed the Nvidia proprietary drivers for Fedora 31i that supports that GPU? It's not uncommon for the non-proprietary drivers to fail to support all the features that FAH needs.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 2:58 am
by GEI
They are Nvidia drivers and Cuda packaged for Fedora. Same as have been using for Tensorflow-gpu
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 3:05 am
by bruce
After installing OpenCL and rebooting, what does the new log say where it previously said
23:56:23: GPUs: 0
23:56:23:CUDA Device 0: Platform:0 Device:0 Bus:5 Slot:0 Compute:6.1 Driver:10.2
23:56:23: OpenCL: Not detected: clGetDeviceIDs() returned -1
The FAHCore does not use CUDA explicitly, it uses OpenCL.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 3:12 am
by GEI
OpenCL is ame as above.
GPU shows now as 1080. Its using the Nvidia GPU under Cuda when checked via nvidia-smi and FAHControl. Seems to need higher amount of cores, but memory use is minimal compared to ML/AI
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 3:19 am
by bruce
FAH uses the CUDA cores by way of the OpenCL API. FAHCore_22 doesn't need large amounts of GPU memory ... just the highly parallel floating point calculation capability (commonly called 3D calculations).
Please post the first couple of pages of FAH's log.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 4:29 pm
by GEI
Will move FAHClient to Docker using a Nvidia base image.
Have to say that while I like the idea of sharing unused resources for research, it needs to be seamless for the volunteer provider.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Tue Mar 10, 2020 5:38 pm
by bruce
GEI wrote:Will move FAHClient to Docker using a Nvidia base image.
Have to say that while I like the idea of sharing unused resources for research, it needs to be seamless for the volunteer provider.
Nobody will disagree with that concept. FAH is mostly scientists thinking about the Science (which is sound) but FAH's code Development resources are very limited.
Re: 1080 Nvidia GPU not available - Fedora 31
Posted: Thu Mar 12, 2020 9:33 pm
by foldy
sudo apt-get install ocl-icd-libopencl1
sudo apt-get install ocl-icd-opencl-dev