B300: FP4 vs FP8 Precision Comparison

FP4 posts 4547 tok/s/GPU for $0.14 per million tokens at 96 tok/s/user on Qwen 3.5 397B-A17B (B300); FP8 posts 1847 tok/s/GPU for $0.36. FP4 is 151% cheaper per token; FP4 delivers 146% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 149 tok/s/user on Qwen 3.5 397B-A17B (B300): FP4 hits 2248 tok/s/GPU, FP8 hits 986. Per-million costs land at $0.29 and $0.65 respectively. FP4 is 126% cheaper per token; FP4 delivers 128% more tok/s/GPU. 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 43–255 tok/s/user interactivity band, at 202 tok/s/user on Qwen 3.5 397B-A17B (B300): FP4 runs 1419 tok/s/GPU at $0.46/M tokens, FP8 runs 662 at $0.98/M. FP4 is 114% cheaper per token; FP4 delivers 115% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks. (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.)