Open MisterHM80 opened 1 year ago
Hi @MisterHM80
could you provide driver version reported by clinfo? I don't know how to map driver version 4000
to our release
Meant between 31.0.101.4032 and 31.0.101.4255 which doesn't work with Async Compute. 31.0.101.3975 or older works as it should.
Or have I misunderstood that?
could you capture API calls using cl intercept? https://github.com/intel/opencl-intercept-layer/blob/main/docs/controls.md
No. I don't see any point in it either. The 5 other tests are running. Just not Async Compute since 4032 version.
Had just installed again 3975 and it works again. I don't need to do anything. Please try to fix it.
@JablonskiMateusz
Any news on this?
Hi, this is now fixed with Intel driver 31.0.101.4369
Thank you for your feedback. I'm glad that's fixed now with Async Compute :)
I hope, the HDR Bug will be fixed soon too That make on 2nd Run slower (Bicubic, Texture and Async Compute), if before run first time HDR. Same for HDR after 2nd Run.
@JablonskiMateusz What about the HDR bug now? Still exists until now. Start the Tool and 1.st run with HDR. 2.nd run is up to 33% slower.
It's about my OpenCL benchmark. Only Windows 64Bit.
"Async Compute" test no longer works since driver version 4xxx. To 3975 went without problems.
With 4xxx version, this only works if I reduce to 512, 512.
Equally buggy runs from the start with "HDR" and "TEXTURE" (Edit : All Driver Version) If you run in order: TEXTURE -> HDR -> TEXTURE -> HDR
Then we see the time indication for each 2nd run higher time.
With Nvidia or AMD there was no such behavior until now.
I would be very grateful if this could finally be fixed. Even if priority is last with OpenCL.
Download (HardwareLuxx Forum) https://www.hardwareluxx.de/community/threads/28-01-2023-r9-fury-added-amd-nvidia-intel-mrh-opencl-benchmark-rankliste.1302003/
Note : The same was already posted on Discord some time ago.