GB300 NVL72: FP4 vs FP8 Precision Comparison

Near the low end of the 19–67 tok/s/user interactivity band, at 31 tok/s/user on GLM 5/5.1 (GB300 NVL72): FP4 runs 9820 tok/s/GPU at $0.07/M tokens, FP8 runs 3824 at $0.19/M. FP4 is 158% cheaper per token; FP4 delivers 157% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 43 tok/s/user on GLM 5/5.1 (GB300 NVL72), FP4 delivers 4835 tok/s/GPU at $0.16 per million tokens; FP8 delivers 276 tok/s/GPU at $2.67. FP4 is 1604% cheaper per token; FP4 delivers 1654% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 849 tok/s/GPU for $0.87 per million tokens at 55 tok/s/user on GLM 5/5.1 (GB300 NVL72); FP8 posts 93 tok/s/GPU for $7.90. FP4 is 813% cheaper per token; FP4 delivers 811% more tok/s/GPU. 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.)