Not enough GPU video memory even with AMD RX 7700 XT

[Isocheims]

I tried to use my AMD RX 7700 XT, which has 12 GB of VRAM, for GPU conversion when using the "MB-Roformer-Inst-V2" MDX-Net model. However, it never works and always pops up an error after a couple of seconds, as shown below. Can anyone figure out why my GPU is showing the low video memory error? Any help is appreciated!

Last Error Received:

Process: MDX-Net

If this error persists, please contact the developers with the error details.

Raw Error Details:

RuntimeError: "Could not allocate tensor with 2327425920 bytes. There is not enough GPU video memory available!"
Traceback Error: "
File "UVR.py", line 7261, in process_start
File "separate.py", line 672, in seperate
File "separate.py", line 845, in demix
File "torch\nn\modules\module.py", line 1501, in _call_impl
File "lib_v5\mel_band_roformer.py", line 442, in forward
File "torch\nn\modules\module.py", line 1501, in _call_impl
File "lib_v5\mel_band_roformer.py", line 160, in forward
File "torch\nn\modules\module.py", line 1501, in _call_impl
File "lib_v5\mel_band_roformer.py", line 121, in forward
File "torch\nn\modules\module.py", line 1501, in _call_impl
File "lib_v5\attend.py", line 95, in forward
File "lib_v5\attend.py", line 74, in flash_attn
"

Error Time Stamp [2024-12-21 17:03:05]

Full Application Settings:

vr_model: Choose Model
aggression_setting: 5
window_size: 512
mdx_segment_size: 256
batch_size: Default
crop_size: 256
is_tta: False
is_output_image: False
is_post_process: False
is_high_end_process: False
post_process_threshold: 0.2
vr_voc_inst_secondary_model: No Model Selected
vr_other_secondary_model: No Model Selected
vr_bass_secondary_model: No Model Selected
vr_drums_secondary_model: No Model Selected
vr_is_secondary_model_activate: False
vr_voc_inst_secondary_model_scale: 0.9
vr_other_secondary_model_scale: 0.7
vr_bass_secondary_model_scale: 0.5
vr_drums_secondary_model_scale: 0.5
demucs_model: Choose Model
segment: Default
overlap: 0.25
overlap_mdx: Default
overlap_mdx23: 8
shifts: 2
chunks_demucs: Auto
margin_demucs: 44100
is_chunk_demucs: False
is_chunk_mdxnet: False
is_primary_stem_only_Demucs: False
is_secondary_stem_only_Demucs: False
is_split_mode: True
is_demucs_combine_stems: True
is_mdx23_combine_stems: True
demucs_voc_inst_secondary_model: No Model Selected
demucs_other_secondary_model: No Model Selected
demucs_bass_secondary_model: No Model Selected
demucs_drums_secondary_model: No Model Selected
demucs_is_secondary_model_activate: False
demucs_voc_inst_secondary_model_scale: 0.9
demucs_other_secondary_model_scale: 0.7
demucs_bass_secondary_model_scale: 0.5
demucs_drums_secondary_model_scale: 0.5
demucs_pre_proc_model: No Model Selected
is_demucs_pre_proc_model_activate: False
is_demucs_pre_proc_model_inst_mix: False
mdx_net_model: MB-Roformer-Inst-v2
chunks: Auto
margin: 44100
compensate: Auto
denoise_option: None
is_match_frequency_pitch: True
phase_option: Automatic
phase_shifts: None
is_save_align: False
is_match_silence: True
is_spec_match: False
is_mdx_c_seg_def: True
is_invert_spec: False
is_deverb_vocals: False
deverb_vocal_opt: Main Vocals Only
voc_split_save_opt: Lead Only
is_mixer_mode: False
mdx_batch_size: Default
mdx_voc_inst_secondary_model: No Model Selected
mdx_other_secondary_model: No Model Selected
mdx_bass_secondary_model: No Model Selected
mdx_drums_secondary_model: No Model Selected
mdx_is_secondary_model_activate: False
mdx_voc_inst_secondary_model_scale: 0.9
mdx_other_secondary_model_scale: 0.7
mdx_bass_secondary_model_scale: 0.5
mdx_drums_secondary_model_scale: 0.5
is_save_all_outputs_ensemble: True
is_append_ensemble_name: False
chosen_audio_tool: Manual Ensemble
choose_algorithm: Min Spec
time_stretch_rate: 2.0
pitch_rate: 2.0
is_time_correction: True
is_gpu_conversion: True
is_primary_stem_only: False
is_secondary_stem_only: False
is_testing_audio: False
is_auto_update_model_params: True
is_add_model_name: False
is_accept_any_input: False
is_save_to_input_path: False
apollo_overlap: 2
apollo_chunk_size: 5
apollo_model: Choose Model
is_task_complete: False
is_normalization: False
is_use_directml: True
is_wav_ensemble: False
is_create_model_folder: False
mp3_bit_set: 320k
semitone_shift: 0
save_format: MP3
wav_type_set: PCM_16
device_set: AMD Radeon RX 7700 XT:0
help_hints_var: True
set_vocal_splitter: No Model Selected
is_set_vocal_splitter: False
is_save_inst_set_vocal_splitter: False
model_sample_mode: False
model_sample_mode_duration: 30
demucs_stems: All Stems
mdx_stems: All Stems

[DrugLover]

are you using opencl version? amd graphic cards are not supported to use torch version

[Isocheims]

I am using the Beta Roformer Version which has DirectML according to the latest release

[deton24]

Go to models\MDX_Net_Models\model_data\mdx_c_configs or look for the yaml corresponding to the model elsewhere if it's not there, and at the end of the config file, edit dim_t to something smaller, e.g. 256 or 201 work on even 4GB VRAM AMD GPUs.
You can set it also in segments while in MDX23C settings, Segment default is disabled, but it might get reset.

If still the same, replace DirectML.dll from your Windows installation into the Ultimate Vocal Remover\torch_directml
but it might cause severe MDX-Net v2 noise with some outdated drivers.

[Isocheims]

Go to models\MDX_Net_Models\model_data\mdx_c_configs or look for the yaml corresponding to the model elsewhere if it's not there, and at the end of the config file, edit dim_t to something smaller, e.g. 256 or 201 work on even 4GB VRAM AMD GPUs. You can set it also in segments while in MDX23C settings, Segment default is disabled, but it might get reset.

If still the same, replace DirectML.dll from your Windows installation into the Ultimate Vocal Remover\torch_directml but it might cause severe MDX-Net v2 noise with some outdated drivers.

Just tried it out and it finally stops crashing! However it is going very, very slowly so I am not sure if it is actually using my GPU or if its my CPU. I have tried out other models besides MB-Roformer-Inst-V2 like Voc_FT and Kim Vocal 2 and they all seem to work way smoother. I think for now I'm going to use a different model as there are less problems. But thanks for the help!