GB300 NVL72: FP4 vs FP8 Precision Comparison

At 71 tok/s/user on DeepSeek R1 (GB300 NVL72), FP4 delivers 11951 tok/s/GPU at $0.06 per million tokens; FP8 delivers 3513 tok/s/GPU at $0.21. FP4 is 241% cheaper per token; FP4 delivers 240% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 4793 tok/s/GPU for $0.15 per million tokens at 128 tok/s/user on DeepSeek R1 (GB300 NVL72); FP8 posts 817 tok/s/GPU for $0.90. FP4 is 484% cheaper per token; FP4 delivers 487% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 184 tok/s/user on DeepSeek R1 (GB300 NVL72): FP4 hits 959 tok/s/GPU, FP8 hits 117. Per-million costs land at $0.81 and $6.40 respectively. FP4 is 693% cheaper per token; FP4 delivers 720% 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.)