Speech-to-text Sourced statement

Whisper small on Mac (Apple Silicon M1–M4)

Benchmark pending. We haven't measured this pair yet. Below is what the sources say about this hardware, quoted with links.

Unified memory shared by CPU and GPU. Runtime support varies a lot (Metal via whisper.cpp/MLX; PyTorch MPS is often partial).

EvidenceSourced statement

Commercial useYes

Licenseapache-2.0

Stated memory~2 GB VRAM

Device memory16 GB unified memory

Results

Benchmark pending. No measured numbers from us yet for Whisper small on Apple Silicon. Get notified or send us your numbers (with the method).

What the sources say about Apple Silicon

Third-party docs whisper.cpp: 'Apple Silicon first-class citizen', optimized via ARM NEON, Accelerate, Metal and Core ML; the Core ML encoder can be more than 3x faster than CPU-only. source
Caveat: No size-specific Mac number published there

Method

We haven't run this pair. Everything on this page is quoted from the linked sources, with the source type labelled (model authors vs third parties). Vendor numbers are often measured under favourable conditions.

Ways to run Whisper small

Runtime	Precision	Documented devices
openai-whisper (PyTorch)	fp16/fp32	CPU, CUDA	source
faster-whisper (CTranslate2)	int8, fp16, fp32	CPU, CUDA	source
whisper.cpp (ggml)	fp16, quantized (q5/q8)	CPU (x86/ARM), Apple Metal + Core ML, CUDA, Vulkan, OpenVINO	source
Hugging Face transformers	fp16/fp32	CPU, CUDA	source

Run it

git clone https://github.com/ggml-org/whisper.cpp && cd whisper.cpp
sh ./models/download-ggml-model.sh small
cmake -B build && cmake --build build -j --config Release   # Metal is used automatically on Apple Silicon
./build/bin/whisper-cli -m models/ggml-small.bin -f samples/jfk.wav

Copied from the linked README. We haven't run it on this device.

Caveats

No size-specific Mac number published there