search

RVC-Project / Retrieval-based-Voice-Conversion-WebUI

Easily train a good VC model with voice data <= 10 mins!
MIT License
19.72k stars 3.05k forks source link

Slow processing in ROCm #2024

Closed Bocchi-Chan2023 closed 2 weeks ago

Bocchi-Chan2023 commented 3 weeks ago

I am not sure if this is the problem, but I will report it in due course. The process of using rmvpe.pt is very slow, it takes 4 minutes (First run took 8 minutes) on the RX6600 while it takes about 1 minute on the GTX1650. Please let me know if there is any information you need. I will report anything. Time: npy: 1.46s, f0: 83.20s, infer: 136.64s.

Kubuntu 22.04 Kernel 6.5.0.28 Ryzen 5 5500 DDR4 3733 32GB rx6600 8GB ROCm 6.0.2

GUUser91 commented 3 weeks ago

I used to have slow processing too on my 7900 xtx. This is how I fixed it. I'm not sure this solution will work for you though. I disable the igpu in the bios. I put in export MIOPEN_FIND_MODE=2 and export HSA_OVERRIDE_GFX_VERSION=11.0.0 in the bashrc file. Then I install the stable build of pytorch.

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.7

There is also a stable build of rocm 6.0. This one also produces fast Inference.

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0