GPU acceleration not working on livecam (RTX 4090)

[mychaylo]

When running the livecam, I just get CPU not GPU. Can't find any discussion on this. Help?

---

Using provider ['CUDAExecutionProvider'] - Device:cuda
Running on local URL: http://127.0.0.1:7861

To create a public link, set share=True in launch().
Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
find model: C:\Users\mfras/.insightface\models\buffalo_l\1k3d68.onnx landmark_3d_68 ['None', 3, 192, 192] 0.0 1.0
Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
find model: C:\Users\mfras/.insightface\models\buffalo_l\2d106det.onnx landmark_2d_106 ['None', 3, 192, 192] 0.0 1.0
Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
find model: C:\Users\mfras/.insightface\models\buffalo_l\det_10g.onnx detection [1, 3, '?', '?'] 127.5 128.0
Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
find model: C:\Users\mfras/.insightface\models\buffalo_l\genderage.onnx genderage ['None', 3, 96, 96] 0.0 1.0
Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
find model: C:\Users\mfras/.insightface\models\buffalo_l\w600k_r50.onnx recognition ['None', 3, 112, 112] 127.5 127.5
set det-size: (640, 640)
Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
inswapper-shape: [1, 3, 128, 128]

[C0untFloyd]

Do you just assume it because it is running so slow (gradio limitation)? It looks like it is using your GPU.

[mychaylo]

yes it could be that, I'm remembering the speed at which it worked in the previous windowed version.

[C0untFloyd]

But previous versions didn't have a livecam (webcam) feature?!? Or are you talking about the preview frame?

[mychaylo]

i guess the preview window? in previous versions of roop there was a live webcam preview window thing, you could capture it with OBS and it ran much faster than this window I see in roop_unleashed. Is there a different method I can use in roop_unleashed? I can't use the other forks of roop because it seems something has been disabled in the downloading (unless there is a workaround there too).

[LuminaDevelopment]

I have a question by pure curiosity: How many frames per second can you perform running face swap using an RTX 4090?

[lysxelapsed]

Depends on a lot of things, like (mostly) which enhancer you use, or your target video resolution.
There's a benchmark thread. Unfortunately, me and C0untFloyd are the only participants so far. It would be nice, if you'd participate too. And by doing it, you can answer your own question and knowing your thread-limit for one, and help others that have the same card.
But my guess would be somewhere around 8-10 frames with enhancer.

[LuminaDevelopment]

okay thanks!