Qwen 3.5 397B-A17B — GB300 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB300 NVL72 (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.
H200 hits 476 tok/s/GPU for $0.82 per million tokens at 53 tok/s/user on Qwen 3.5 397B-A17B. No GB300 NVL72 data at this operating point.
H200: 323 tok/s/GPU, $1.22 per million tokens at 80 tok/s/user on Qwen 3.5 397B-A17B. GB300 NVL72 is unmeasured here.
At 106 tok/s/user on Qwen 3.5 397B-A17B, H200 delivers 272 tok/s/GPU at $1.44 per million tokens; GB300 NVL72 hasn't been benchmarked at this target. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:—H200:475.6 | GB300 NVL72:—H200:322.8 | GB300 NVL72:—H200:272.5 |
| Cost ($/M tok) | GB300 NVL72:—H200:$0.821 | GB300 NVL72:—H200:$1.216 | GB300 NVL72:—H200:$1.438 |
| tok/s/MW | GB300 NVL72:—H200:274884 | GB300 NVL72:—H200:186573 | GB300 NVL72:—H200:157493 |
| Concurrency | GB300 NVL72:—H200:~37 | GB300 NVL72:—H200:~17 | GB300 NVL72:—H200:~12 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.