MI355X: FP4 vs FP8 Precision Comparison

At 66 tok/s/user on DeepSeek R1 (MI355X), FP4 delivers 1848 tok/s/GPU at $0.22 per million tokens; FP8 delivers 807 tok/s/GPU at $0.50. FP4 is 121% cheaper per token; FP4 delivers 129% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 642 tok/s/GPU for $0.64 per million tokens at 116 tok/s/user on DeepSeek R1 (MI355X); FP8 posts 295 tok/s/GPU for $1.41. FP4 is 119% cheaper per token; FP4 delivers 118% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 166 tok/s/user on DeepSeek R1 (MI355X): FP4 hits 206 tok/s/GPU, FP8 hits 130. Per-million costs land at $1.99 and $3.59 respectively. FP4 is 80% cheaper per token; FP4 delivers 59% more tok/s/GPU. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data. (Numbers reflect the default 1k/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.)