GB300 NVL72: FP4 vs FP8 Precision Comparison

Near the low end of the 19–115 tok/s/user interactivity band, at 43 tok/s/user on GLM 5/5.1 (GB300 NVL72): FP4 runs 10120 tok/s/chip at $0.06/M tokens, FP8 runs 2618 at $0.25/M. FP4 is 287% cheaper per token; FP4 delivers 287% more tok/s/chip. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 67 tok/s/user on GLM 5/5.1 (GB300 NVL72), FP4 delivers 6960 tok/s/chip at $0.09 per million tokens; FP8 delivers 479 tok/s/chip at $1.34. FP4 is 1353% cheaper per token; FP4 delivers 1353% more tok/s/chip. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 3619 tok/s/chip for $0.18 per million tokens at 91 tok/s/user on GLM 5/5.1 (GB300 NVL72); FP8 posts 171 tok/s/chip for $3.74. FP4 is 2011% cheaper per token; FP4 delivers 2011% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab. (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.)