B200: FP4 vs FP8 Precision Comparison

Throughput at 95 tok/s/user on Qwen 3.5 397B-A17B (B200): FP4 hits 5845 tok/s/GPU, FP8 hits 1931. Per-million costs land at $0.09 and $0.28 respectively. FP4 is 206% cheaper per token; FP4 delivers 203% 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 31–286 tok/s/user interactivity band, at 159 tok/s/user on Qwen 3.5 397B-A17B (B200): FP4 runs 2193 tok/s/GPU at $0.25/M tokens, FP8 runs 876 at $0.61/M. FP4 is 149% cheaper per token; FP4 delivers 150% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 223 tok/s/user on Qwen 3.5 397B-A17B (B200), FP4 delivers 995 tok/s/GPU at $0.54 per million tokens; FP8 delivers 536 tok/s/GPU at $0.98. FP4 is 82% cheaper per token; FP4 delivers 86% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact. (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.)