Comfy-Org/comfy-aimdo

Fork 0

mirror of https://github.com/Comfy-Org/comfy-aimdo.git synced 2026-11-04 23:37:21 +00:00

[GH-ISSUE #95] [ROCM/Linux]: VBAR allocation fails on RDNA1 #33

New issue

Open

opened 2026-10-26 22:23:50 +00:00 by bsakaguchi · 6 comments

bsakaguchi commented

2026-10-26 22:23:50 +00:00

Owner

Originally created by @Zakhrov on GitHub (Aug 25, 2026).
Original GitHub issue: https://github.com/Comfy-Org/comfy-aimdo/issues/95

Stacktrace with a clean ComfyUI install:

 uv run main.py[INFO] setup plugin alembic.autogenerate.schemas
[INFO] setup plugin alembic.autogenerate.tables
[INFO] setup plugin alembic.autogenerate.types
[INFO] setup plugin alembic.autogenerate.constraints
[INFO] setup plugin alembic.autogenerate.defaults
[INFO] setup plugin alembic.autogenerate.comments
[INFO] setup plugin alembic.autogenerate.checkconstraint_byname
[INFO] Found comfy_kitchen backend hip: {'available': False, 'disabled': False, 'unavailable_reason': 'architecture is not in the validated target manifest: gfx1010, gfx90c', 'capabilities': []}
[INFO] Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'int8_linear', 'na3d', 'quantize_and_rotate_rowwise', 'quantize_int8_rowwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'w4a8_int8_linear']}
[INFO] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_embedding', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'int8_linear', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']}
[INFO] Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']}
[INFO] Checkpoint files will always be loaded safely.
[INFO] Total VRAM 6128 MB, total RAM 63664 MB
[INFO] pytorch version: 2.12.0+rocm10.1.0a20260824
[INFO] AMD arch: gfx1010
[INFO] ROCm version: (7, 16)
[INFO] Set vram state to: NORMAL_VRAM
[INFO] Device: cuda:0 AMD Radeon RX 5600M : native
[INFO] Device: cuda:1 AMD Radeon Graphics : native
[INFO] Using async weight offloading with 2 streams
[INFO] Enabled pinned memory 57297.0
[INFO] Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
[INFO] aimdo: /project/src/control.c:276:INFO:comfy-aimdo inited for GPU: AMD Radeon RX 5600M (VRAM: 6128 MB)
[INFO] aimdo: /project/src/control.c:276:INFO:comfy-aimdo inited for GPU: AMD Radeon Graphics (VRAM: 32768 MB)
[INFO] DynamicVRAM support detected and enabled
[INFO] Python version: 3.14.5 (main, May 10 2026, 19:28:16) [Clang 22.1.3 ]
[INFO] ComfyUI version: 0.33.0
[INFO] comfy-aimdo version: 0.4.13
[INFO] comfy-kitchen version: 0.2.31
[INFO] comfyui-frontend-package version: 1.49.6
[INFO] comfyui-workflow-templates version: 0.11.46
[INFO] comfyui-embedded-docs version: 0.5.10
[INFO] comfy-kitchen version: 0.2.31
[INFO] comfy-aimdo version: 0.4.13
[INFO] [Prompt Server] web root: /home/aaron/Projects/personal/clean_comfy/ComfyUI/.venv/lib/python3.14/site-packages/comfyui_frontend_package/static
[INFO] Asset seeder disabled
[INFO] No OpenGL_accelerate module loaded: No module named 'OpenGL_accelerate'
[INFO] 
Import times for custom nodes:
[INFO]    0.0 seconds: /home/aaron/Projects/personal/clean_comfy/ComfyUI/custom_nodes/websocket_image_save.py
[INFO] 
[INFO] Context impl SQLiteImpl.
[INFO] Will assume non-transactional DDL.
[INFO] Using RAM pressure cache.
[INFO] Starting server

[INFO] To see the GUI go to: http://127.0.0.1:8188
[INFO] got prompt
[INFO] model weight dtype torch.float16, manual cast: None
[INFO] model_type EPS
[INFO] Using split attention in VAE
[INFO] Using split attention in VAE
[INFO] VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
[INFO] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[INFO] Requested to load SD1ClipModel
[ERROR] aimdo: /project/src/model-vbar.c:236:ERROR:Could not reseve Virtual Address space for VBAR
[ERROR] !!! Exception during processing !!! VBAR allocation failed
[ERROR] Traceback (most recent call last):
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 545, 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, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 344, 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, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 318, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 306, in process_inputs
    result = f(**inputs)
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/nodes.py", line 77, in encode
    return (clip.encode_from_tokens_scheduled(tokens), )
            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/sd.py", line 341, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/sd.py", line 405, in encode_from_tokens
    self.load_model(tokens)
    ~~~~~~~~~~~~~~~^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/sd.py", line 463, in load_model
    model_management.load_models_gpu([self.patcher], memory_required=memory_used)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_management.py", line 1004, in load_models_gpu
    loaded_model.model_load(lowvram_model_memory, force_patch_weights=force_patch_weights)
    ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_management.py", line 790, in model_load
    self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)
    ~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_management.py", line 818, in model_use_more_vram
    return self.model.partially_load(self.device, extra_memory, force_patch_weights=force_patch_weights)
           ~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 2153, in partially_load
    raise e
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 2150, in partially_load
    self.load(device_to, dirty=dirty)
    ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 1873, in load
    vbar = self._vbar_get(create=True)
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 1805, in _vbar_get
    vbar = comfy_aimdo.model_vbar.ModelVBAR(self.model_size() * 10, self.load_device.index)
  File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/.venv/lib/python3.14/site-packages/comfy_aimdo/model_vbar.py", line 54, in __init__
    raise MemoryError("VBAR allocation failed")
MemoryError: VBAR allocation failed

[INFO] Prompt executed in 0.22 seconds

Enabling cudamallocasync has no effect

Originally created by @Zakhrov on GitHub (Aug 25, 2026). Original GitHub issue: https://github.com/Comfy-Org/comfy-aimdo/issues/95 Stacktrace with a clean ComfyUI install: ``` uv run main.py[INFO] setup plugin alembic.autogenerate.schemas [INFO] setup plugin alembic.autogenerate.tables [INFO] setup plugin alembic.autogenerate.types [INFO] setup plugin alembic.autogenerate.constraints [INFO] setup plugin alembic.autogenerate.defaults [INFO] setup plugin alembic.autogenerate.comments [INFO] setup plugin alembic.autogenerate.checkconstraint_byname [INFO] Found comfy_kitchen backend hip: {'available': False, 'disabled': False, 'unavailable_reason': 'architecture is not in the validated target manifest: gfx1010, gfx90c', 'capabilities': []} [INFO] Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'int8_linear', 'na3d', 'quantize_and_rotate_rowwise', 'quantize_int8_rowwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'w4a8_int8_linear']} [INFO] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_embedding', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'int8_linear', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']} [INFO] Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']} [INFO] Checkpoint files will always be loaded safely. [INFO] Total VRAM 6128 MB, total RAM 63664 MB [INFO] pytorch version: 2.12.0+rocm10.1.0a20260824 [INFO] AMD arch: gfx1010 [INFO] ROCm version: (7, 16) [INFO] Set vram state to: NORMAL_VRAM [INFO] Device: cuda:0 AMD Radeon RX 5600M : native [INFO] Device: cuda:1 AMD Radeon Graphics : native [INFO] Using async weight offloading with 2 streams [INFO] Enabled pinned memory 57297.0 [INFO] Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention [INFO] aimdo: /project/src/control.c:276:INFO:comfy-aimdo inited for GPU: AMD Radeon RX 5600M (VRAM: 6128 MB) [INFO] aimdo: /project/src/control.c:276:INFO:comfy-aimdo inited for GPU: AMD Radeon Graphics (VRAM: 32768 MB) [INFO] DynamicVRAM support detected and enabled [INFO] Python version: 3.14.5 (main, May 10 2026, 19:28:16) [Clang 22.1.3 ] [INFO] ComfyUI version: 0.33.0 [INFO] comfy-aimdo version: 0.4.13 [INFO] comfy-kitchen version: 0.2.31 [INFO] comfyui-frontend-package version: 1.49.6 [INFO] comfyui-workflow-templates version: 0.11.46 [INFO] comfyui-embedded-docs version: 0.5.10 [INFO] comfy-kitchen version: 0.2.31 [INFO] comfy-aimdo version: 0.4.13 [INFO] [Prompt Server] web root: /home/aaron/Projects/personal/clean_comfy/ComfyUI/.venv/lib/python3.14/site-packages/comfyui_frontend_package/static [INFO] Asset seeder disabled [INFO] No OpenGL_accelerate module loaded: No module named 'OpenGL_accelerate' [INFO] Import times for custom nodes: [INFO] 0.0 seconds: /home/aaron/Projects/personal/clean_comfy/ComfyUI/custom_nodes/websocket_image_save.py [INFO] [INFO] Context impl SQLiteImpl. [INFO] Will assume non-transactional DDL. [INFO] Using RAM pressure cache. [INFO] Starting server [INFO] To see the GUI go to: http://127.0.0.1:8188 [INFO] got prompt [INFO] model weight dtype torch.float16, manual cast: None [INFO] model_type EPS [INFO] Using split attention in VAE [INFO] Using split attention in VAE [INFO] VAE load device: cuda:0, offload device: cpu, dtype: torch.float32 [INFO] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] Requested to load SD1ClipModel [ERROR] aimdo: /project/src/model-vbar.c:236:ERROR:Could not reseve Virtual Address space for VBAR [ERROR] !!! Exception during processing !!! VBAR allocation failed [ERROR] Traceback (most recent call last): File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 545, 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, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 344, 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, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 318, in _async_map_node_over_list await process_inputs(input_dict, i) File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/execution.py", line 306, in process_inputs result = f(**inputs) File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/nodes.py", line 77, in encode return (clip.encode_from_tokens_scheduled(tokens), ) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/sd.py", line 341, in encode_from_tokens_scheduled pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True) File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/sd.py", line 405, in encode_from_tokens self.load_model(tokens) ~~~~~~~~~~~~~~~^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/sd.py", line 463, in load_model model_management.load_models_gpu([self.patcher], memory_required=memory_used) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_management.py", line 1004, in load_models_gpu loaded_model.model_load(lowvram_model_memory, force_patch_weights=force_patch_weights) ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_management.py", line 790, in model_load self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights) ~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_management.py", line 818, in model_use_more_vram return self.model.partially_load(self.device, extra_memory, force_patch_weights=force_patch_weights) ~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 2153, in partially_load raise e File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 2150, in partially_load self.load(device_to, dirty=dirty) ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 1873, in load vbar = self._vbar_get(create=True) File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/comfy/model_patcher.py", line 1805, in _vbar_get vbar = comfy_aimdo.model_vbar.ModelVBAR(self.model_size() * 10, self.load_device.index) File "/home/aaron/Projects/personal/clean_comfy/ComfyUI/.venv/lib/python3.14/site-packages/comfy_aimdo/model_vbar.py", line 54, in __init__ raise MemoryError("VBAR allocation failed") MemoryError: VBAR allocation failed [INFO] Prompt executed in 0.22 seconds ``` Enabling cudamallocasync has no effect

bsakaguchi commented

2026-10-26 22:23:51 +00:00

Author

Owner

@0xDELUXA commented on GitHub (Aug 25, 2026):

I’m not sure RDNA1 is on the support roadmap.

@0xDELUXA commented on GitHub (Aug 25, 2026): I’m not sure RDNA1 is on the support roadmap.

bsakaguchi commented

2026-10-26 22:23:52 +00:00

Author

Owner

@Zakhrov commented on GitHub (Aug 25, 2026):

I’m not sure RDNA1 is on the support roadmap.

rocm 7.15 and later has support for it with a few caveats like triton and flash attention not working yet. Running it with dynamic vram disabled and using alternatives like ggufs and distributed torch nodes works for now.

@Zakhrov commented on GitHub (Aug 25, 2026): > I’m not sure RDNA1 is on the support roadmap. rocm 7.15 and later has support for it with a few caveats like triton and flash attention not working yet. Running it with dynamic vram disabled and using alternatives like ggufs and distributed torch nodes works for now.

bsakaguchi commented

2026-10-26 22:23:52 +00:00

Author

Owner

@0xDELUXA commented on GitHub (Aug 25, 2026):

rocm 7.15 and later has support for it with a few caveats like triton and flash attention not working yet. Running it with dynamic vram disabled and using alternatives like ggufs and distributed torch nodes works for now.

I see. Good to know, but I meant support in ComfyUI.

If Triton doesn't work, then FlashAttention or SageAttention isn't possible either. What error does Triton give on RDNA1? Also, int8 convrot or w4a8 is a much better quantization type than GGUF. And what do you mean by "distributed torch nodes"?

@0xDELUXA commented on GitHub (Aug 25, 2026): > > rocm 7.15 and later has support for it with a few caveats like triton and flash attention not working yet. Running it with dynamic vram disabled and using alternatives like ggufs and distributed torch nodes works for now. I see. Good to know, but I meant support in ComfyUI. If Triton doesn't work, then FlashAttention or SageAttention isn't possible either. What error does Triton give on RDNA1? Also, int8 convrot or w4a8 is a much better quantization type than GGUF. And what do you mean by "distributed torch nodes"?

bsakaguchi commented

2026-10-26 22:23:52 +00:00

Author

Owner

@Zakhrov commented on GitHub (Aug 25, 2026):

rocm 7.15 and later has support for it with a few caveats like triton and flash attention not working yet. Running it with dynamic vram disabled and using alternatives like ggufs and distributed torch nodes works for now.

I see. Good to know, but I meant support in ComfyUI.

If Triton doesn't work, then FlashAttention or SageAttention isn't possible either. What error does Triton give on RDNA1? Also, int8 convrot or w4a8 is a much better quantization type than GGUF. And what do you mean by "distributed torch nodes"?

Triton doesn't give an error, it just outputs gray garbled images on anything newer than SD 1.5 based checkpoints/models. int8 convrot and w4a8 don't have native acceleration, and they upcast to fp16 while hitting OOM errors in certain cases.
The workaround I've been using is the ComfyUI-MultiGPU node pack that allows virtual memory offloading using Distributed Torch 2.

@Zakhrov commented on GitHub (Aug 25, 2026): > > rocm 7.15 and later has support for it with a few caveats like triton and flash attention not working yet. Running it with dynamic vram disabled and using alternatives like ggufs and distributed torch nodes works for now. > > I see. Good to know, but I meant support in ComfyUI. > > If Triton doesn't work, then FlashAttention or SageAttention isn't possible either. What error does Triton give on RDNA1? Also, int8 convrot or w4a8 is a much better quantization type than GGUF. And what do you mean by "distributed torch nodes"? Triton doesn't give an error, it just outputs gray garbled images on anything newer than SD 1.5 based checkpoints/models. int8 convrot and w4a8 don't have native acceleration, and they upcast to fp16 while hitting OOM errors in certain cases. The workaround I've been using is the [ComfyUI-MultiGPU](https://github.com/pollockjj/ComfyUI-MultiGPU) node pack that allows virtual memory offloading using Distributed Torch 2.

bsakaguchi commented

2026-10-26 22:23:52 +00:00

Author

Owner

@Zakhrov commented on GitHub (Aug 26, 2026):

int8_convrot doesn't work with RDNA1 it throws:

CUDA error: HIPBLAS_STATUS_INVALID_VALUE when calling `hipblasLtMatmulAlgoGetHeuristic( ltHandle, computeDesc.descriptor(), Adesc.descriptor(), Bdesc.descriptor(), Cdesc.descriptor(), Cdesc.descriptor(), preference.descriptor(), 1, &heuristicResult, &returnedResult)`

@Zakhrov commented on GitHub (Aug 26, 2026): int8_convrot doesn't work with RDNA1 it throws: ``` CUDA error: HIPBLAS_STATUS_INVALID_VALUE when calling `hipblasLtMatmulAlgoGetHeuristic( ltHandle, computeDesc.descriptor(), Adesc.descriptor(), Bdesc.descriptor(), Cdesc.descriptor(), Cdesc.descriptor(), preference.descriptor(), 1, &heuristicResult, &returnedResult)` ```

bsakaguchi commented

2026-10-26 22:23:52 +00:00

Author

Owner

@0xDELUXA commented on GitHub (Aug 26, 2026):

That's what I said about RDNA1 support in ComfyUI.

@0xDELUXA commented on GitHub (Aug 26, 2026): That's what I said about RDNA1 support in ComfyUI.

bsakaguchi referenced this issue

2026-10-26 22:24:20 +00:00

[PR #36] [MERGED] Add aarch64/ARM64 Linux build support #75