I was doing a bit of troubleshooting on the Windoze box last night and discovered MLC requires a fairly new driver for Nvidia cards. I was using 445.75, and it works but seemed to have some stability problems for me. I went back to 418.91 which works well for other projects, but is too old for MLC. Right now the latest certified/recommended driver for Windows is 452.06. I haven't tried it yet.
MLC@Home Nvidia GPU App
- Thread starter Nick Name
- Start date
Note at MLC.
"AMD GPU support
ROCm support is here, but there are quire a few caveats. The biggest is that it only supports VEGA-based discrete cards at the moment, which means VEGA56/64, Radeon VII, Several MIxxx cards. NAVI1 and NAVI2 aren't fully supported by ROCm, POLARIS discrete graphics support is (temporarily) broken in ROCm 3.9, and should be fixed when ROCm 4.0 is released (was announced last week). When ROCm 4.0 is released I'll re-spin the rocm client with that which should enable POLARIS-based GPU crunching as well (RX 550,560,570,580,590). APUs are not supported. Windows is not supported.
Currently, the server will only serve ROCm WUs to machines with "Radeon RX Vega" in their host/os id string and (once we implement a bug fix to the boinc core code) is running kernel 5.0.0 or greater.
At the moment, ROCm support is slower than CUDA, but still faster than CPU."
"AMD GPU support
ROCm support is here, but there are quire a few caveats. The biggest is that it only supports VEGA-based discrete cards at the moment, which means VEGA56/64, Radeon VII, Several MIxxx cards. NAVI1 and NAVI2 aren't fully supported by ROCm, POLARIS discrete graphics support is (temporarily) broken in ROCm 3.9, and should be fixed when ROCm 4.0 is released (was announced last week). When ROCm 4.0 is released I'll re-spin the rocm client with that which should enable POLARIS-based GPU crunching as well (RX 550,560,570,580,590). APUs are not supported. Windows is not supported.
Currently, the server will only serve ROCm WUs to machines with "Radeon RX Vega" in their host/os id string and (once we implement a bug fix to the boinc core code) is running kernel 5.0.0 or greater.
At the moment, ROCm support is slower than CUDA, but still faster than CPU."
I'm getting this on BOINC log all day using GTX 760 with 451.48 driver in Win 7. Sounds like site not driver. But I could try newer driver. Any ideas?
11/24/2020 11:47:50 PM | MLC@Home | work fetch resumed by user
11/24/2020 11:47:56 PM | MLC@Home | Sending scheduler request: To fetch work.
11/24/2020 11:47:56 PM | MLC@Home | Requesting new tasks for NVIDIA GPU
11/24/2020 11:47:57 PM | | Project communication failed: attempting access to reference site
11/24/2020 11:47:57 PM | MLC@Home | Scheduler request failed: Couldn't resolve host name
11/24/2020 11:47:57 PM | MLC@Home | Sending scheduler request: To fetch work.
11/24/2020 11:47:57 PM | MLC@Home | Requesting new tasks for NVIDIA GPU
11/24/2020 11:47:59 PM | | Internet access OK - project servers may be temporarily down.
11/24/2020 11:47:59 PM | MLC@Home | Scheduler request completed: got 0 new tasks
11/24/2020 11:47:59 PM | MLC@Home | Project requested delay of 31 seconds
11/24/2020 11:47:50 PM | MLC@Home | work fetch resumed by user
11/24/2020 11:47:56 PM | MLC@Home | Sending scheduler request: To fetch work.
11/24/2020 11:47:56 PM | MLC@Home | Requesting new tasks for NVIDIA GPU
11/24/2020 11:47:57 PM | | Project communication failed: attempting access to reference site
11/24/2020 11:47:57 PM | MLC@Home | Scheduler request failed: Couldn't resolve host name
11/24/2020 11:47:57 PM | MLC@Home | Sending scheduler request: To fetch work.
11/24/2020 11:47:57 PM | MLC@Home | Requesting new tasks for NVIDIA GPU
11/24/2020 11:47:59 PM | | Internet access OK - project servers may be temporarily down.
11/24/2020 11:47:59 PM | MLC@Home | Scheduler request completed: got 0 new tasks
11/24/2020 11:47:59 PM | MLC@Home | Project requested delay of 31 seconds
ROCm seems to be the future for AMD GPU compute. I'm not sure what that means for OpenCL.
Free-DC status page now says MLC is down.
ROCm provides the OpenCL support. Once installed it can be verified with the following commands:
/opt/rocm/bin/rocminfo
/opt/rocm/opencl/bin/clinfo
Since I have a machine(actually a few) that I can't move to other things due to disk space limitations, I have been debating trying to install a GPU in them since they are PCIe 4.0 capable. Have also been debating whether to follow the AMD's CDNA (compute oriented architecture (VEGA)) or RDNA (Graphics oriented architecture (NAVI)). It looks like CDNA is the data center compute platform with HBM2 memory and cards cost $1,000 up. Radeon VII might be the exception but the cards seem to be scarce. Unlike Virtualbox (Oracle), AMD hasn't been able to completely grasp open source on Linux. The result is you have to completely uninstall their code before doing upgrades to anything which can be a pain in the backside. After the first of the year I may get one RX6800 XT and physically install it then see if the current level of AMDGPU recognizes the card. Then after playing with just the graphics part, I may try to install ROCm and see if it can do compute functions. Supposedly, one doesn't need the proprietary driver but I'm thinking that AMD is usually a little behind in the open source support so things might not work right away on newer products.
/opt/rocm/bin/rocminfo
/opt/rocm/opencl/bin/clinfo
Since I have a machine(actually a few) that I can't move to other things due to disk space limitations, I have been debating trying to install a GPU in them since they are PCIe 4.0 capable. Have also been debating whether to follow the AMD's CDNA (compute oriented architecture (VEGA)) or RDNA (Graphics oriented architecture (NAVI)). It looks like CDNA is the data center compute platform with HBM2 memory and cards cost $1,000 up. Radeon VII might be the exception but the cards seem to be scarce. Unlike Virtualbox (Oracle), AMD hasn't been able to completely grasp open source on Linux. The result is you have to completely uninstall their code before doing upgrades to anything which can be a pain in the backside. After the first of the year I may get one RX6800 XT and physically install it then see if the current level of AMDGPU recognizes the card. Then after playing with just the graphics part, I may try to install ROCm and see if it can do compute functions. Supposedly, one doesn't need the proprietary driver but I'm thinking that AMD is usually a little behind in the open source support so things might not work right away on newer products.
I had some time tonight and tried upgrading to 452.06. MLC did run, I didn't run any tasks to completion as the GPU load was basically nil. I aborted that work and 15-20 minutes later the driver crashed taking GPUGrid and Amicable Number work with it. There are mixed reports on the MLC forum about GPU work, some have decent results but others see what I did. It's not worth it for me especially since the app is listed as Win10 and I'm not upgrading this system to that.
There isn't a working Linux GPU app yet but I'm going to stick to CPU even if one is available. I hate messing with working drivers, I'm back to 418.91 on the Windoze box.