AMD VEGA APU Support

[fighter3005]

I was wondering if anyone has been able to run Comfy on an APU like the 5600G or similar?
I have rocm 6.3 with HWE Kernel 6.11 on Ubuntu and pytorch for rocm 6.3 installed. I set

export HSA_OVERRIDE_GFX_VERSION=9.0.6
export PYTORCH_ROCM_ARCH="gfx906"

since the APU is not supported I think!? I always get

CUDA error: HIPBLAS_STATUS_INTERNAL_ERROR when calling hipblasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)

I have an NVidia driver and gpu installed, but it should not be used (ideally)

I have the same problem with ForgeUI. And I don't get why? I really thought this approach should work.
Does anyone had the same problem and can provide some insight?

Thanks!

Checkpoint files will always be loaded safely.
Total VRAM 31834 MB, total RAM 63668 MB
pytorch version: 2.7.1+rocm6.3
AMD arch: gfx906:sramecc-:xnack+
ROCm version: (6, 3)
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon Graphics : native
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
Python version: 3.10.13 (main, Apr  6 2025, 08:23:24) [GCC 13.3.0]
ComfyUI version: 0.3.44
ComfyUI frontend version: 1.23.4

To see the GUI go to: http://0.0.0.0:8188
To see the GUI go to: http://[::]: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 SD1ClipModel
loaded completely 30582.03828125 235.84423828125 True
!!! Exception during processing !!! CUDA error: HIPBLAS_STATUS_INTERNAL_ERROR when calling `hipblasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)`
Traceback (most recent call last):
  File "/home/user/coding/ComfyUI/execution.py", line 425, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
  File "/home/user/coding/ComfyUI/execution.py", line 268, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
  File "/home/user/coding/ComfyUI/execution.py", line 242, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "/home/user/coding/ComfyUI/execution.py", line 230, in process_inputs
    result = f(**inputs)
  File "/home/user/coding/ComfyUI/nodes.py", line 69, in encode
    return (clip.encode_from_tokens_scheduled(tokens), )
  File "/home/user/coding/ComfyUI/comfy/sd.py", line 167, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
  File "/home/user/coding/ComfyUI/comfy/sd.py", line 229, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
  File "/home/user/coding/ComfyUI/comfy/sd1_clip.py", line 683, in encode_token_weights
    out = getattr(self, self.clip).encode_token_weights(token_weight_pairs)
  File "/home/user/coding/ComfyUI/comfy/sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
  File "/home/user/coding/ComfyUI/comfy/sd1_clip.py", line 288, in encode
    return self(tokens)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/sd1_clip.py", line 261, in forward
    outputs = self.transformer(None, attention_mask_model, embeds=embeds, num_tokens=num_tokens, intermediate_output=intermediate_output, final_layer_norm_intermediate=self.layer_norm_hidden_state, dtype=torch.float32)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/clip_model.py", line 145, in forward
    x = self.text_model(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/clip_model.py", line 118, in forward
    x, i = self.encoder(x, mask=mask, intermediate_output=intermediate_output)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/clip_model.py", line 70, in forward
    x = l(x, mask, optimized_attention)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/clip_model.py", line 52, in forward
    x += self.mlp(self.layer_norm2(x))
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/clip_model.py", line 37, in forward
    x = self.fc1(x)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/user/.pyenv/versions/comfy-amd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/ops.py", line 84, in forward
    return self.forward_comfy_cast_weights(*args, **kwargs)
  File "/home/user/coding/ComfyUI/comfy/ops.py", line 80, in forward_comfy_cast_weights
    return torch.nn.functional.linear(input, weight, bias)
RuntimeError: CUDA error: HIPBLAS_STATUS_INTERNAL_ERROR when calling `hipblasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)`