HIP error: unspecified launch failure

[remvage]

After 4-5 generations using rocm on 9070xt (CPU 5800x3D, no integrated graphics, AMD_SERIALIZE_KERNEL=3 used at startup) in Windows 11, errors such as the following occur: 2025-12-20 12:17:48, 174 | remvage-pc | sd | ERROR | processing_helpers | Torch generator: seeds=[1840411946] device=cuda HIP error: unspecified launch failure
Search for hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

Only a complete PC restart helps. Sometimes this error does not occur, even after generating 100-150 images. Has anyone else encountered this?

log:

HIP error: unspecified launch failure
Search for hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

2025-12-20 12:17:45,861 | remvage-pc | sd | ERROR | errors | Processing: AcceleratorError
2025-12-20 12:17:45,980 | remvage-pc | sd | DEBUG | sd_checkpoint | Search model: name="novaAnimeXL_ilV140 [aaa94191eb]" matched="C:\sdnext\models\Stable-diffusion\novaAnimeXL_ilV140.safetensors" type=alias
2025-12-20 12:17:46,002 | remvage-pc | sd | DEBUG | modelstats | Analyzed: model="novaAnimeXL_ilV140" type=sdxl class=StableDiffusionXLPipeline size=6938436084 mtime="2025-12-18 17:17:08" modules=[name="vae" cls=AutoencoderKL config=True device=cpu dtype=torch.bfloat16 params=83653863 modules=243, name="text_encoder" cls=CLIPTextModel config=True device=cpu dtype=torch.bfloat16 params=123060480 modules=152, name="text_encoder_2" cls=CLIPTextModelWithProjection config=True device=cpu dtype=torch.bfloat16 params=694659840 modules=393, name="tokenizer" cls=CLIPTokenizer config=False, name="tokenizer_2" cls=CLIPTokenizer config=False, name="unet" cls=UNet2DConditionModel config=True device=cuda:0 dtype=torch.bfloat16 params=2567463684 modules=1930, name="scheduler" cls=EulerAncestralDiscreteScheduler config=True, name="image_encoder" cls=NoneType config=False, name="feature_extractor" cls=NoneType config=False, name="force_zeros_for_empty_prompt" cls=bool config=False]
2025-12-20 12:17:46,006 | remvage-pc | sd | INFO | processing_diffusers | Processing modifiers: unapply
2025-12-20 12:17:46,018 | remvage-pc | sd | DEBUG | processing | Process: batch=1/1 interrupted
2025-12-20 12:17:46,019 | remvage-pc | sd | INFO | processing | Processed: images=0 its=0.00 ops=['txt2img']
2025-12-20 12:17:46,020 | remvage-pc | sd | DEBUG | processing | Processed: timers={'total': 0.2, 'post': 0.19}
2025-12-20 12:17:46,022 | remvage-pc | sd | DEBUG | processing | Processed: memory={'ram': {'total': 31.93, 'rss': 17.24, 'used': 24.89, 'free': 7.04, 'avail': 7.04, 'buffers': 0, 'cached': 0}, 'gpu': {'used': 0, 'total': 0, 'active': 0.4, 'peak': 0.57, 'retries': 0, 'oom': 0, 'swap': 0, 'error': "HIP error: unspecified launch failure\nSearch for hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information.\nHIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\nFor debugging consider passing AMD_SERIALIZE_KERNEL=3\nCompile with TORCH_USE_HIP_DSAto enable device-side assertions.\n"}, 'job': ''} 2025-12-20 12:17:46,339 | remvage-pc | sd | DEBUG | devices | GC: current={'gpu': 0, 'ram': 24.89, 'oom': 0} prev={'gpu': 0, 'ram': 24.89} load={'gpu': 0, 'ram': 78} gc={'gpu': 0, 'py': 157} fn=process_images:process_images_inner why=final time=0.31 2025-12-20 12:17:46,348 | remvage-pc | sd | DEBUG | ui_control_helpers | Select input: type=<class 'str'> source=None init=None mask=None mode=Изображение time=0.00 2025-12-20 12:17:47,457 | remvage-pc | sd | DEBUG | ui_control_helpers | Select input: type=<class 'str'> source=None init=None mask=None mode=Изображение time=0.00 2025-12-20 12:17:48,153 | remvage-pc | sd | DEBUG | sd_samplers | Sampler: "Default" cls=EulerAncestralDiscreteScheduler config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'trained_betas': None, 'prediction_type': 'epsilon', 'timestep_spacing': 'trailing', 'steps_offset': 1, 'rescale_betas_zero_snr': False, 'interpolation_type': 'linear', 'use_karras_sigmas': False, 'clip_sample': False, 'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True} 2025-12-20 12:17:48,155 | remvage-pc | sd | INFO | processing_diffusers | Processing modifiers: apply 2025-12-20 12:17:48,168 | remvage-pc | sd | ERROR | prompt_parser_diffusers | Prompt encode: HIP error: unspecified launch failure Search forhipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information.
HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing AMD_SERIALIZE_KERNEL=3
Compile with TORCH_USE_HIP_DSA to enable device-side assertions.

2025-12-20 12:17:48,170 | remvage-pc | sd | WARNING | processing_args | Prompt parser encode: empty prompt embeds
2025-12-20 12:17:48,171 | remvage-pc | sd | ERROR | prompt_parser_diffusers | Prompt encode: HIP error: unspecified launch failure
Search for hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

2025-12-20 12:17:48,172 | remvage-pc | sd | ERROR | prompt_parser_diffusers | Prompt encode: HIP error: unspecified launch failure
Search for hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

2025-12-20 12:17:48,174 | remvage-pc | sd | ERROR | processing_helpers | Torch generator: seeds=[1840411946] device=cuda HIP error: unspecified launch failure
Search for hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

[vladmandic]

english only please

[remvage]

corrected

[remvage]

An error occurred again with the following text. Perhaps some additional information about the system is required?

23:15:02-075208 DEBUG Processed: memory={'ram': {'total': 31.93, 'rss': 17.01, 'used': 24.97, 'free': 6.96, 'avail':
6.96, 'buffers': 0, 'cached': 0}, 'gpu': {'used': 0, 'total': 0, 'active': 0.24, 'peak': 1.93,
'retries': 0, 'oom': 0, 'swap': 0, 'error': "HIP error: unspecified launch failure\nSearch for
hipErrorLaunchFailure' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__HIPRT__TYPES.html for more information.\nHIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\nFor debugging consider passing AMD_SERIALIZE_KERNEL=3\nCompile with TORCH_USE_HIP_DSA` to enable device-side assertions.\n"},
'job': ''}
23:15:02-424786 DEBUG GC: current={'gpu': 0, 'ram': 24.97, 'oom': 0} prev={'gpu': 0, 'ram': 24.97} load={'gpu': 0,
'ram': 78} gc={'gpu': 0, 'py': 56402} fn=process_images:process_images_inner why=final
time=0.35
23:15:02-424786 ERROR Control exception: 'oom'
23:15:02-424786 ERROR Control: KeyError
╭───────────────────────────────────────── Traceback (most recent call last) ──────────────────────────────────────────╮
│C:\sdnext\modules\ui_control.py:105 in generate_click │
│ │
│ 104 │ │ │ │ progress.record_results(job_id, results) │
│❱ 105 │ │ │ │ yield return_controls(results, t) │
│ 106 │ │ except Exception as e: │
│ │
│C:\sdnext\modules\ui_control.py:54 in return_controls │
│ │
│ 53 │ else: │
│❱ 54 │ │ perf = return_stats(t) │
│ 55 │ if res is None: # no response │
│ │
│C:\sdnext\modules\ui_control.py:32 in return_stats │
│ │
│ 31 │ │ mem_mon_read = shared.mem_mon.read() │
│❱ 32 │ │ ooms = mem_mon_read.pop("oom") │
│ 33 │ │ retries = mem_mon_read.pop("retries") │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
KeyError: 'oom'