Vulkan performance on windows compared to linux is terrible.
Windows - 700ms GPU, 1200ms CPU vs Linux - 52ms GPU on the same file. tested on 3070 & 3090 with similar gap in results
I understand you removed cuda due to size constraints but it would be nice to just dump those on google drive or give instructions to compile ourselves.
For my use case the gpu timing difference is night and day. idk where you got the results in the release for 1.4.0 but i can't get anywhere near those results even when just cloning and running the repo.
EDIT
Recompiled from whisper.cpp 1.7.5 and it fixed it. now getting 100ms on windows on a 3070. seems like there is something wrong with the existing dll files.
Vulkan performance on windows compared to linux is terrible.
Windows - 700ms GPU, 1200ms CPU vs Linux - 52ms GPU on the same file. tested on 3070 & 3090 with similar gap in results
I understand you removed cuda due to size constraints but it would be nice to just dump those on google drive or give instructions to compile ourselves.
For my use case the gpu timing difference is night and day. idk where you got the results in the release for 1.4.0 but i can't get anywhere near those results even when just cloning and running the repo.
EDIT
Recompiled from whisper.cpp 1.7.5 and it fixed it. now getting 100ms on windows on a 3070. seems like there is something wrong with the existing dll files.