ROCm: [Issue]: Comfy_UI hipblasLT not supported for Radeon 7900XT and HSA_STATUS_ERROR_OUT_OF_REGISTERS error

Problem Description

I’m reporting this issue on behalf of the OP from this post on discuss.pytorch.org. I’ve copied the stdout/stderr here

Total VRAM 20464 MB, total RAM 63432 MB
pytorch version: 2.6.0.dev20241122+rocm6.2
Set vram state to: NORMAL_VRAM
Device: cuda:0 Radeon RX 7900 XT : native
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
[Prompt Server] web root: /home/sersys/pyproj/ComfyUI-0.3.12/web

Import times for custom nodes:
   0.0 seconds: /home/sersys/pyproj/ComfyUI-0.3.12/custom_nodes/websocket_image_save.py

Starting server

To see the GUI go to: http://127.0.0.1:8188
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using split attention in VAE
Using split attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
Requested to load SDXLClipModel
loaded completely 9.5367431640625e+25 1560.802734375 True
/home/sersys/pyproj/ComfyUI-0.3.12/comfy/ops.py:64: UserWarning: Attempting to use hipBLASLt on an unsupported architecture! Overriding blas backend to hipblas (Triggered internally at /pytorch/aten/src/ATen/Context.cpp:296.)
  return torch.nn.functional.linear(input, weight, bias)
Requested to load SDXL
loaded completely 9.5367431640625e+25 4897.0483474731445 True
  0%|                                                                | 0/20 [00:00<?, ?it/s]:0:rocdevice.cpp            :2984: 93862815501 us: [pid:318562 tid:0x7f122e5ff640] Callback: Queue 0x7f0ed8000000 aborting with error : HSA_STATUS_ERROR_OUT_OF_REGISTERS: Kernel has requested more VGPRs than are available on this agent code: 0x2d
Aborted (core dumped)

There are two issues here that the OP is concerned about :

The first is a non-fatal warning There appears to be a warning issued by pytorch stating that the Radeon RX 7900XT is not supported by the HIPBlasLT backend. There is no ROCm documentation that indicates this GPU is not supported by HIPBlasLT; in fact the compatibility matrix suggests it is supported. Is this an issue with Pytorch or is this GPU not actually supported in hipblaslt ?

This being said, it does appear that they are not on a supported Linux kernel ( 6.12.9 ); however, no errors appear to indicate specifically that this is the culprit.

The second issue appears fatal During the execution of torch.nn.functional.linear, the code aborts with an HSA_STATUS_ERROR_OUT_OF_REGISTERS error. I suspect this an issue with MIOpen; perhaps it’s targeting the wrong architecture and over-allocating vgpr ? My understanding is that some of these torch functions are compiled on the fly during the first execution and this is handled by MIOpen. Any guidance you can provide on further diagnosing the problem here would be great.

Operating System

Debian GNU/Linux 12

CPU

AMD Ryzen 9 7900X 12-Core Processor

GPU

Radeon 7900 XT

ROCm Version

ROCm 6.2.0

ROCm Component

Pytorch, MIOpen

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 32 (1 by maintainers)

Most upvoted comments

Apologies, I’ve gotten this mixed up with other Comfy_UI issues we’re tracking. This particular one is on Debian.

Understandable. Thank you for everyone’s help. I appreciate you all taking your time for me.

Ok this worked. uname -a

Linux sy-lu 6.11.0-19-generic #19~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Mon Feb 17 11:51:52 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

This is on a separate hard drive, no distrobox, no container, just as is. Other info: python3 -V

Python 3.12.3

rocminfo rocomubuntu.txt

pip show torch

pip show torch
Name: torch
Version: 2.8.0.dev20250314+rocm6.3
Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
Home-page: https://pytorch.org/
Author: PyTorch Team
Author-email: packages@pytorch.org
License: BSD-3-Clause
Location: /home/sersys/pyproj/torch/lib/python3.12/site-packages
Requires: filelock, fsspec, jinja2, networkx, pytorch-triton-rocm, setuptools, sympy, typing-extensions
Required-by: kornia, spandrel, torchaudio, torchsde, torchvision

Comfy_ui version v0.3.26

I have actually generated my first image.

Thanks @tcgu-amd for testing that out. I’ll relay this back to OP