My TrueNAS Scale system
Asrock Rack ROME2D16-2T motherboard
2x AMD EPYC 7742 64-core
256GB (16x 16GB) DDR4-3200 RDIMM
RTX 3050 -> passed through to Plex VM
RTX 2080Ti -> used by host/apps
storage/pools not really relevant for this conversation.
So I know there are many different hands at play for my specific issue, so please bear with me. I'll explain the whole situation as best I can and hopefully someone can narrow down if this is a TrueNAS issue, a TrueCharts issue, or the container maintainer issue.
One of the things I use this system for is BOINC. TrueCharts offers a BOINC container, which as I understand it is just a kind of port or translation (sorry if I misunderstand what exactly they do here) of the BOINC docker image provided by LSIO. I was previously using this BOINC chart no problem with just CPU access.
for those that aren't aware of what BOINC is, it's a distributed computing platform that allows donating compute time and resources to various scientific research projects. I'm VERY well versed in BOINC itself and how it works and operates on Linux, but I haven't used it in a container much.
Now here's the problem. when the container starts, it recognizes the GPU fine, but ONLY for CUDA. no OpenCL detected (listed in the BOINC Event Log at startup). which means that I cannot run any OpenCL Nvidia applications. some projects only have OpenCL Nvidia applications and do not use CUDA. the tasks from projects that do use CUDA work fine, they are using the GPU normally and I can see the tasks running in nvidia-smi, they complete and get submitted, etc.
I know that by default, the standard Nvidia Linux drivers do not include working OpenCL. you normally have to install a second package (ocl-icd-libopencl1) to get OpenCL support in Linux/Debian. I'm not sure how I can verify if this package is or is not installed in TrueNAS Scale. if this package is not installed in Scale, then that might be the source of the problem. can anyone verify this? or recommend some commands I can run on the host to see if OpenCL works? normally I would install a package called clinfo and check the detected ocl platforms, but I'm not really able to do this with the appliance type nature of TrueNAS.
or maybe it's a way that TrueCharts sets up the container? I'm not sure how to verify anything with that though.
or maybe it's a problem in the base docker image from LSIO itself? again, not sure how to verify that.
any help on where to start troubleshooting would be appreciated. I'd like to try to narrow down where the issue actually exists to avoid the spider-man meme of everyone pointing the blame at someone else.
Asrock Rack ROME2D16-2T motherboard
2x AMD EPYC 7742 64-core
256GB (16x 16GB) DDR4-3200 RDIMM
RTX 3050 -> passed through to Plex VM
RTX 2080Ti -> used by host/apps
storage/pools not really relevant for this conversation.
So I know there are many different hands at play for my specific issue, so please bear with me. I'll explain the whole situation as best I can and hopefully someone can narrow down if this is a TrueNAS issue, a TrueCharts issue, or the container maintainer issue.
One of the things I use this system for is BOINC. TrueCharts offers a BOINC container, which as I understand it is just a kind of port or translation (sorry if I misunderstand what exactly they do here) of the BOINC docker image provided by LSIO. I was previously using this BOINC chart no problem with just CPU access.
for those that aren't aware of what BOINC is, it's a distributed computing platform that allows donating compute time and resources to various scientific research projects. I'm VERY well versed in BOINC itself and how it works and operates on Linux, but I haven't used it in a container much.
Now here's the problem. when the container starts, it recognizes the GPU fine, but ONLY for CUDA. no OpenCL detected (listed in the BOINC Event Log at startup). which means that I cannot run any OpenCL Nvidia applications. some projects only have OpenCL Nvidia applications and do not use CUDA. the tasks from projects that do use CUDA work fine, they are using the GPU normally and I can see the tasks running in nvidia-smi, they complete and get submitted, etc.
I know that by default, the standard Nvidia Linux drivers do not include working OpenCL. you normally have to install a second package (ocl-icd-libopencl1) to get OpenCL support in Linux/Debian. I'm not sure how I can verify if this package is or is not installed in TrueNAS Scale. if this package is not installed in Scale, then that might be the source of the problem. can anyone verify this? or recommend some commands I can run on the host to see if OpenCL works? normally I would install a package called clinfo and check the detected ocl platforms, but I'm not really able to do this with the appliance type nature of TrueNAS.
or maybe it's a way that TrueCharts sets up the container? I'm not sure how to verify anything with that though.
or maybe it's a problem in the base docker image from LSIO itself? again, not sure how to verify that.
any help on where to start troubleshooting would be appreciated. I'd like to try to narrow down where the issue actually exists to avoid the spider-man meme of everyone pointing the blame at someone else.
