B200: FP4 vs FP8 Precision Comparison

FP4 posts 9367 tok/s/chip for $0.05 per million tokens at 77 tok/s/user on DeepSeek R1 (B200); FP8 posts 2297 tok/s/chip for $0.21. FP4 is 308% cheaper per token; FP4 delivers 308% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 138 tok/s/user on DeepSeek R1 (B200): FP4 hits 2267 tok/s/chip, FP8 hits 989. Per-million costs land at $0.21 and $0.49 respectively. FP4 is 129% cheaper per token; FP4 delivers 129% more tok/s/chip. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data.

Toward the upper edge of the 16–260 tok/s/user interactivity band, at 199 tok/s/user on DeepSeek R1 (B200): FP4 runs 1409 tok/s/chip at $0.34/M tokens, FP8 runs 362 at $1.33/M. FP4 is 289% cheaper per token; FP4 delivers 289% more tok/s/chip. 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.)