Benchmark script output comparison:
=== KERNEL 7.0.0-28 ===
Start: 2026-07-24 02:32:40 CST
Kernel: 7.0.0-28-generic
PyTorch: 2.14.0+rocm7.15.0
[1/4] Loading checkpoint file from disk...
Done in 0.0s
[2/4] Parsing model (CLIP, VAE, UNET)...
<<< HUNG HERE - NO OUTPUT FOR 300 SECONDS >>>
End: 2026-07-24 02:37:42 CST (5 minute timeout)
Stack trace from Ctrl+C at same step:
File "comfy/ops.py", line 477, in _load_from_state_dict
param.copy_(input_param)
File "comfy/sd.py", line 274, in __init__
m, u = self.load_sd(state_dict, full_model=True)
[repeated 7 more times through torch nn module load_state_dict]
=== KERNEL 6.17.0-35 ===
Start: 2026-07-24 02:42:58 CST
Kernel: 6.17.0-35-generic
PyTorch: 2.14.0+rocm7.15.0
[1/4] Loading checkpoint file from disk...
Done in 0.0s
[2/4] Parsing model (CLIP, VAE, UNET)...
Done in 5.4s
[3/4] Moving main model to GPU...
(proceeded to next step normally)
End: 2026-07-24 02:43:08 CST
Same PyTorch, same ROCm, same model, same ComfyUI version.
Only difference: kernel version.
Step 2 (CLIP weight loading) is 55x+ slower on kernel 7.0 due to KFD work queue
deadlock.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2161667
Title:
Kernel 7.0 KFD deadlock loading models via ROCm on Strix Halo
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-hwe-7.0/+bug/2161667/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
