** Description changed: BugLink: https://bugs.launchpad.net/bugs/2170529 [ Impact ] gfx1103 covers both GC 11.0.1 (PHX1/HPT1) and GC 11.0.4 (PHX2/HPT2). There are known stability issues with ROCm on with gfx1103 that are root - caused to the GC firmware[1]. The updated firmware resolves these - issues. + caused to the GC firmware. + + There's a PyTorch test failure reported publicly[1]. From our internal + testing this may also manifest as GPU hang when running workload with + vLLM and ComfyUI. The updated firmware resolves these issues. [ Fix ] Update linux-firmware binaries for gfx1103 to at least commit 9ec083c3aba9[2] [ Test Plan ] 1. For all releases: boot and run basic graphic tests (e.g. glxgears, glmark2). Make sure graphic functions work. 2. Additionally, for a release where ROCm is meant to support: re-run the test in [1], and make sure that doesn't reproduce. [ Where problems could occur] Bad firmware may impact compute workload and graphic rendering on the said silicon. However because of how amdgpu works, this only impacts platforms with the said IP blocks. Test packages of linux-firmware-amd-graphics including these firmware were built for Noble, Resolute, Stonking and wrere tested internally by our team, and they confirmed the firmware is working on the respective releases. [ Additional information] [1] https://github.com/ROCm/TheRock/issues/6252 [2] https://gitlab.com/kernel-firmware/linux-firmware/-/commit/9ec083c3aba932190e3c160e52cbb7e264148b61
-- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2170529 Title: [SRU] Update GC firmware for gfx1103 To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux-firmware-amd-graphics/+bug/2170529/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
