The author explores the performance of the Whisper Benchmark with the MLX Framework from Apple, comparing it to the Nvidia 4090 and other Apple Silicon models. They provide code and results from their benchmarking tests, highlighting the similarity in structure between the Apple and Python whisper frameworks. The author finds that the Nvidia 4090 is still faster than the Apple Silicon models, but notes that the new M3 Max Chip with more GPU cores could potentially surpass it. They also emphasize that the measurements are not scientifically accurate but provide a rough idea of the capabilities of the MLX framework. The author shares their motivation for conducting these tests, which is to improve their podcast search engine.
https://owehrens.com/whisper-nvidia-rtx-4090-vs-m1pro-with-mlx/