Qwen 3.5 397B-A17B · GPU comparison

Qwen 3.5 397B-A17B — B300 vs H200

Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on Qwen 3.5 397B-A17B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.

Throughput at 68 tok/s/user on Qwen 3.5 397B-A17B: B300 hits 2918 tok/s/GPU, H200 hits 490. Per-million costs land at $0.22 and $0.80 respectively. B300 is 262% cheaper per token; B300 delivers 495% more tok/s/GPU.

B300 / H200 on Qwen 3.5 397B-A17B at 106 tok/s/user: 1559 / 351 tok/s/GPU, $0.42 / $1.10 per million tokens. B300 is 165% cheaper per token; B300 delivers 344% more tok/s/GPU.

Toward the upper edge of the 30–182 tok/s/user interactivity band, at 144 tok/s/user on Qwen 3.5 397B-A17B: B300 runs 1036 tok/s/GPU at $0.63/M tokens, H200 runs 284 at $1.37/M. B300 is 120% cheaper per token; B300 delivers 264% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)

View performance-per-dollar view →

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/gpu)
B300:2918.4H200:490.2
B300:1559.2H200:351.0
B300:1035.7H200:284.4
Cost ($/M tok)
B300:$0.221H200:$0.799
B300:$0.416H200:$1.102
B300:$0.626H200:$1.374
tok/s/MW
B300:1344867H200:283348
B300:718509H200:202875
B300:477260H200:164386
Concurrency
B300:~87H200:~29
B300:~31H200:~13
B300:~15H200:~8

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.

Vendor:
Aggregation:
Spec Decoding: