mirror of
https://github.com/Const-me/Whisper.git
synced 2026-10-18 13:10:52 +08:00
Konstantin
30 lines
1.7 KiB
Plaintext
30 lines
1.7 KiB
Plaintext
This folder contains 2 sample speech audio clips, `jfk.wav` and `columbia.wma`
|
||
|
||
jfk.wav is from whisper.cpp repository.
|
||
|
||
columbia.wma is from Wikipedia: https://upload.wikimedia.org/wikipedia/commons/1/1f/George_W_Bush_Columbia_FINAL.ogg
|
||
I re-encoded the audio from Ogg Vorbis into Windows Media Audio, because Media Foundation is unable to decode Vorbis.
|
||
|
||
The rest of the text files in this folder are the outputs of the in-app performance profiler, when the app was transcribing these two audio clips on three different computers.
|
||
|
||
The “1080ti” files are from my desktop, which has nVidia GeForce 1080Ti GPU.
|
||
|
||
The “vega8” files are from the same desktop, when using the GPU integrated into AMD Ryzen 7 5700G processor.
|
||
|
||
The “vega7” files are from my laptop, the GPU is integrated into AMD Ryzen 5 5600U processor.
|
||
The laptop model is HP ProBook 445 G8. While running the tests, the laptop was on battery power.
|
||
|
||
The “1650” files are from another desktop with nVidia GeForce 1650.
|
||
|
||
The file names with “medium” in the middle were made with “ggml-medium.bin” Whisper model, with “large” were made with “ggml-large.bin” model.
|
||
|
||
In theory, 1080ti delivers 10.6 TFlops FP32 and 484.4 GB/second VRAM bandwidth.
|
||
|
||
The AMD APU in that desktop delivers 2.0 TFlops FP32, and 53.3 GB/second memory bandwidth.
|
||
|
||
That variant of 1650 delivers 2.6 TFlops FP32, and 128.1 GB/second VRAM bandwidth.
|
||
|
||
The AMD APU in the laptop delivers 1.6 TFlops FP32, and 51.2 GB/second memory bandwidth.
|
||
|
||
The loading times in these logs make little sense because they depend heavily on the disk cache implemented in the OS kernel.
|
||
Ideally, I needed to reboot Windows before every test, but I didn’t do that. |