Whisper turbo vs. medium: Speed ranking flips between CPU and GPU on two machines
A developer benchmarked whisper.cpp v1.9.5 on four builds across two PCs, one with an x86 CPU and Intel integrated GPU and one with an Arm CPU and NVIDIA RTX Spark CUDA GPU. Using the same short Japanese recording and arguments, medium was faster on both CPUs, while turbo was faster on CUDA and roughly tied on the integrated GPU once totals were compared. Per-phase timings showed turbo gaining in the decoder and losing in the encoder, with the balance depending on hardware. The author also found that an MSVC build on Windows on Arm lacked NEON and ran 5.6 to 10.2 times slower than a clang build, and that model loading dominated total time on faster GPUs.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in