Qwen 3.5 397B-A17B · B300 · Precision comparison

B300: FP4 vs BF16 precision comparison

How FP4 and BF16 precision affect Qwen 3.5 397B-A17B inference on B300 (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 102 tok/s/user on Qwen 3.5 397B-A17B (B300), FP4 delivers 4,220 tok/s/chip at $0.15 per million tokens; BF16 delivers 1,366 tok/s/chip at $0.46. FP4 delivers 3.1× the tok/s/chip on the same chip, which makes it 68% cheaper per token. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 2,057 tok/s/chip for $0.31 per million tokens at 160 tok/s/user on Qwen 3.5 397B-A17B (B300); BF16 posts 712 tok/s/chip for $0.88. FP4 delivers 189% more tok/s/chip on the same chip, which makes it 65% cheaper per token. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 218 tok/s/user on Qwen 3.5 397B-A17B (B300): FP4 hits 1,242 tok/s/chip, BF16 hits 466. Costs per million tokens land at $0.51 and $1.35 respectively. FP4 delivers 167% more tok/s/chip on the same chip, which makes it 62% 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)
FP44,220BF161,366
FP42,057BF16712
FP41,242BF16466
Cost ($/M tok)
FP4$0.15BF16$0.46
FP4$0.31BF16$0.88
FP4$0.51BF16$1.35
tok/s/MW
FP42,220,925BF16718,720
FP41,082,881BF16374,720
FP4653,919BF16245,270
Concurrency
FP4~64BF16~63
FP4~16BF16~20
FP4~12BF16~5
Qwen 3.5 397B-A17B: FP4 versus BF16 throughput and cost at matched interactivity levels on B300
FP4 versus BF16 throughput and cost per million tokens on B300 for this comparison's canonical default workload
Open full-size image ↗