You are not logged in.
Hi everyone, not sure in which section this belongs so I tried Applications.
I want to get my GPU (Vega 64) running fah using rocm. Due to the Wiki this should be supported. It shows correctly in the config:
******************************* System ********************************
19:24:07: CPU: AMD Ryzen 7 1700X Eight-Core Processor
19:24:07: CPU ID: AuthenticAMD Family 23 Model 1 Stepping 1
19:24:07: CPUs: 16
19:24:07: Memory: 15.63GiB
19:24:07: Free Memory: 10.80GiB
19:24:07: Threads: POSIX_THREADS
19:24:07: OS Version: 5.5
19:24:07: Has Battery: false
19:24:07: On Battery: false
19:24:07: UTC Offset: 1
19:24:07: PID: 2982
19:24:07: CWD: /opt/fah
19:24:07: OS: Linux 5.5.10-arch1-1 x86_64
19:24:07: OS Arch: AMD64
19:24:07: GPUs: 0
19:24:07: CUDA: Not detected: Failed to open dynamic library 'libcuda.so':
19:24:07: libcuda.so: cannot open shared object file: No such file or
19:24:07: directory
19:24:07:OpenCL Device 0: Platform:0 Device:0 Bus:13 Slot:0 Compute:2.0 Driver:3084.0
19:24:07:***********************************************************************When the client tries to get a new packet for the GPU it immediately stops. The Log gives the following error:
BAD_WORK_UNIT / Error initializing context: clCreateCommandQueue (-6)Any Ideas?
System:
Kernel Version: 5.5.10-arch1-1
foldingathome: 7.5.1-1
rocm-opencl-runtime: 3.1.0-1
Last edited by theringoffire (2020-03-25 10:07:31)
Offline
Same problem here. I could proceed a step by installing the AUR package opencl-amd, which was suggested in the fah forum. The opencl-amd package uses parts of the AMDGPU-PRO package. Actually I thought this is not necessary and fah would run with opencl-mesa, but unfortunately it did not.
However fah still doesn't work. Now I'm stuck with a kernel crash that happens each time fah wants to start processing the work unit. There's no error in fah's log file and the fah process seems stuck, it can only be stopped with "kill -9".
I'll try it again when amd-mesa when mesa 20.0 is in stable.
Offline
I have a patch on the dri mailing list that should fix the OpenCL issues on 5.5 or 5.6rc kernels
Offline
Now with kernel 5.6 and the latest opencl-amd package fah finally works for me. No BAD_WORK_UNIT and no crashes anymore.
@theringoffire: There's also a new rocm release (v3.3) and I've seen reports stating that fah works with it.
Offline
This is why I marked the issue as SOLVED.
Offline
I'm having the same problem
BAD_WORK_UNIT / Error initializing context: clCreateCommandQueue (-6) I have a Radeon R4 integrated and a dedicated Radeon R5 m240.
The integrated folds OK, the R5 fails.
I'm using opencl-amd and kernel 5.6.8
Thanks
Last edited by Ignacc (2020-05-04 04:49:37)
Offline