So far, I got the best luck with amdgpu-install_22.20.50201-1_all.deb.
Background:
I recently got a pair of RX 6400 (OEM: XFX Speedster SWFT105), just to check out. The cost was below an GTX 1650 and it only uses 53W compared to 75W or more (depending on model). All the other GPUs I have are different nVidia ranging from GTX 1650 to RTX A2000, including some M class Tesla and Quadro. The slowest are GTX 1650 (TU116) and according to techpowerup.com's theoretical data, the RX 6400 have better performance than the GTX 1650. All my computer hardware are involved in BOINC projects.
In a number of attempts, I have been trying to get OpenCL to work, but failed every time. The BOINC software has not been able to detect the two RX 6400 properly. BOINC has an issue, looking for the outdated libaticalrt.so, but that's not problematic. The second best attempt was with amdgpu-install_22.20.50200-1438747~22.04_all.deb. BOINC accepted then the GPUs, but every WU (task) failed initially. Checking stderr files of concern, the gfx1030-amdgcn-mesa-mesa3d.bc file was reported to be missing. Checked the files on the computer and it was correct, but it's the wrong file. RX 6400 require gtx1034-... The GPU has been reported as "Beige Goby" only. (Probably nothing to bother about, but still an indication of improper work)
Last and best attempt. Found amdgpu-install_22.20.50201-1_all.deb, installed it and tested. The gfx1030 and Beige Goby issues was fixed. This time, the GPUs started to work, but failed at a later stage. In BOINC, I tested with work units (WU) from PrimeGRID and Einstein. With Einstein, it failed after just 13 seconds. The related stderr.txt starts with
process exited with code 69 (0x45, -187)
and ends with
17:39:28 (2816): [debug]: Flags: X64 SSE SSE2 GNUC X86 GNUX86
17:39:28 (2816): [debug]: glibc version/release: 2.35/stable
17:39:28 (2816): [debug]: Set up communication with graphics process.
boinc_get_opencl_ids returned [0x1e062c0 , 0x7f35eeb42d90]
Using OpenCL platform provided by: Advanced Micro Devices, Inc.
Using OpenCL device "gfx1034" by: Advanced Micro Devices, Inc.
Max allocation limit: 3636461568
Global mem size: 4278190080
Couldn't create OpenCL command queue (error: -6)!
OpenCL shutdown complete!
initialize_ocl returned error [2013]
OCL context null
OCL queue null
Error generating generic FFT context object [5]
17:39:28 (2816): [CRITICAL]: ERROR: MAIN() returned with error '5'
FPU status flags:
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out.cohfu': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out.cohfu': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out.cohfu': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out.cohfu': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out.cohfu': No such file or directory
mv: cannot stat 'LATeah3012L12220719_812.0_0_0.0_3194361_0_0.out.cohfu': No such file or directory
17:39:40 (2816): [normal]: done. calling boinc_finish(69).
17:39:40 (2816): called boinc_finish
The rest in between are only parameters used by the cruncher.
With PrimeGRID, I got
process exited with code 1 (0x1, -255)</message>
<stderr_txt>
AP26 OpenCL 10-shift search version 3.1 by Bryan Little and Iain Bethune
Compiled Aug 7 2020 with GCC 7.5.0
Command line: ../../projects/www.primegrid.com/ap27_3.01_opencl_linux64 267839065 267839184 0
Error: Creating Command Queue. (clCreateCommandQueue) returned -6
</stderr_txt>
The crunchers in both project reports basically the same. A OpenCL issue.
Here, with the same subproject, I noticed that it's slow, very slow. When the GTX 1650 was finished with one task, the RX 6400 that started at the very same time, had only completed about 10%. When the RX 6400 finally reached just above 90% completion, the crunch speed dropped to about a 1/10th. The first of the two RX 6400 who reached just above 97%, caused the other one to fail too, regardless of progress.
As I have spent hours trying to make this work, I soon will reach a limit. After passing that point, I will sell the two RX 6400 to someone that doesn't care about OpenCL and buy a RTX A2000, despite way higher cost (maybe wait and buy a RTX A4000). With nVidia, I will immediately get working Plug and Play ...
