https://bugzilla.kernel.org/show_bug.cgi?id=221297
Bug ID: 221297
Summary: AMDGPU SMU driver interface version mismatch on R9700
- fan control broken under load
Product: Drivers
Version: 2.5
Hardware: All
OS: Linux
Status: NEW
Severity: normal
Priority: P3
Component: Video(DRI - non Intel)
Assignee: [email protected]
Reporter: [email protected]
Regression: No
Hardware: ASUS Turbo Radeon AI Pro R9700 32GB
vBIOS: 115-G287BP00-100
OS: Ubuntu 24.04
Kernel: 6.17.0-19
ROCm: 7.2.1
AMDGPU driver: 6.16.13
BUG:
The GPU fan does not spin up automatically under thermal load. During AI
training the GPU reached 109°C and thermally throttled with the fan physically
stationary throughout.
ROOT CAUSE - SMU interface version mismatch (from dmesg):
amdgpu 0000:2d:00.0: amdgpu: smu driver if version = 0x0000002e (46)
amdgpu 0000:2d:00.0: amdgpu: smu fw if version = 0x00000032 (50)
amdgpu 0000:2d:00.0: amdgpu: smu fw version = 0x00684b00 (104.75.0)
amdgpu 0000:2d:00.0: amdgpu: SMU driver if version not matched
The card firmware is 4 interface versions ahead of what the AMDGPU driver
supports. As a result all fan control registers are inaccessible:
- rocm-smi --setfan returns 'Not supported on this system'
- sysfs pwm1 node is READ-ONLY (-r--r--r--)
- fan1_enable returns 'Invalid argument' when read
- GPU enters runtime power suspend under load, further suppressing fan response
Confirmed on AMDGPU 6.16.6 and 6.16.13 under ROCm 7.2.1.
FIX REQUIRED:
The amdgpu kernel driver needs to be updated to support SMU interface version
50 (0x00000032) as shipped on the R9700 (gfx1201, RDNA 4).
Related ROCm issue: https://github.com/ROCm/ROCm/issues/5908
--
You may reply to this email to add a comment.
You are receiving this mail because:
You are watching the assignee of the bug.
