MI355X: FP4 vs FP8 Precision Comparison

FP4 posts 6547 tok/s/GPU for $0.06 per million tokens at 2 tok/s/user on DeepSeek V4 Pro 1.6T (MI355X); FP8 posts 25 tok/s/GPU for $16.2. FP4 is 25741% cheaper per token; FP4 delivers 25753% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 2 tok/s/user on DeepSeek V4 Pro 1.6T (MI355X): FP4 hits 6547 tok/s/GPU, FP8 hits 25. Per-million costs land at $0.06 and $16.2 respectively. FP4 is 25741% cheaper per token; FP4 delivers 25753% more tok/s/GPU. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data.

Around the middle of the 2–2 tok/s/user interactivity band, at 2 tok/s/user on DeepSeek V4 Pro 1.6T (MI355X): FP4 runs 6547 tok/s/GPU at $0.06/M tokens, FP8 runs 25 at $16.2/M. FP4 is 25741% cheaper per token; FP4 delivers 25753% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks. (Numbers reflect the default 8k/1k selection for this URL — table and chart below update if you change sequence or model in the controls. Each side uses the best available serving configuration for that precision, which may include speculative decoding such as MTP where recipes exist — the same convention as the other comparison pages.)