GLM 5/5.1 · B200 · Precision comparison

B200: FP4 vs FP8 precision comparison

How FP4 and FP8 precision affect GLM 5/5.1 inference on B200 (NVIDIA Blackwell). Throughput, latency, and cost across LLM workloads. Use the chart controls below to switch sequences and metrics — same interactions as the main inference chart.

At 46 tok/s/user on GLM 5/5.1 (B200), FP4 delivers 2,935 tok/s/chip at $0.16 per million tokens; FP8 delivers 1,381 tok/s/chip at $0.35. FP4 delivers 113% more tok/s/chip on the same chip, which makes it 53% cheaper per token. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 1,980 tok/s/chip for $0.24 per million tokens at 78 tok/s/user on GLM 5/5.1 (B200); FP8 posts 906 tok/s/chip for $0.53. FP4 delivers 119% more tok/s/chip on the same chip, which makes it 54% cheaper per token. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 110 tok/s/user on GLM 5/5.1 (B200): FP4 hits 1,417 tok/s/chip, FP8 hits 569. Costs per million tokens land at $0.34 and $0.84 respectively. FP4 delivers 149% more tok/s/chip on the same chip, which makes it 60% cheaper per token. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data. (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.)

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.

Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/chip)
FP42,935FP81,381
FP41,980FP8906
FP41,417FP8569
Cost ($/M tok)
FP4$0.16FP8$0.35
FP4$0.24FP8$0.53
FP4$0.34FP8$0.84
tok/s/MW
FP41,716,344FP8807,393
FP41,157,761FP8529,549
FP4828,931FP8332,578
Concurrency
FP4~29FP8~28
FP4~12FP8~11
FP4~6FP8~5
GLM 5/5.1: FP4 versus FP8 throughput and cost at matched interactivity levels on B200
FP4 versus FP8 throughput and cost per million tokens on B200 for this comparison's canonical default workload
Open full-size image ↗