Llama 3.3 70B · B200 · Precision comparison

B200: FP4 vs FP8 precision comparison

How FP4 and FP8 precision affect Llama 3.3 70B 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 76 tok/s/user on Llama 3.3 70B (B200), FP4 delivers 5,971 tok/s/chip at $0.080 per million tokens; FP8 delivers 4,431 tok/s/chip at $0.11. FP4 delivers 35% more tok/s/chip on the same chip, which makes it 26% cheaper per token. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 3,099 tok/s/chip for $0.16 per million tokens at 108 tok/s/user on Llama 3.3 70B (B200); FP8 posts 2,744 tok/s/chip for $0.18. FP4 delivers 13% more tok/s/chip on the same chip, which makes it 11% cheaper per token. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 139 tok/s/user on Llama 3.3 70B (B200): FP4 hits 1,032 tok/s/chip, FP8 hits 912. Costs per million tokens land at $0.47 and $0.53 respectively. FP4 delivers 13% more tok/s/chip on the same chip, which makes it 12% 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 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.)

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)
FP45,971FP84,431
FP43,099FP82,744
FP41,032FP8912
Cost ($/M tok)
FP4$0.080FP8$0.11
FP4$0.16FP8$0.18
FP4$0.47FP8$0.53
tok/s/MW
FP43,491,881FP82,591,472
FP41,812,036FP81,604,732
FP4603,441FP8533,421
Concurrency
FP4~84FP8~128
FP4~63FP8~54
FP4~16FP8~16
Llama 3.3 70B: 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 ↗