Qwen 3.5 397B-A17B · Chip comparison
Qwen 3.5 397B-A17B — GB300 NVL72 vs H100
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H100 (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.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX →
GB300 NVL72 / H100 on Qwen 3.5 397B-A17B at 108 tok/s/user: 73,544 / 4,154 tok/s/chip, $0.009 / $0.078 per million tokens. GB300 NVL72 is 797% cheaper per token; GB300 NVL72 delivers 1670% more tok/s/chip.
Around the middle of the 68–228 tok/s/user interactivity band, at 148 tok/s/user on Qwen 3.5 397B-A17B: GB300 NVL72 runs 57,478 tok/s/chip at $0.011/M tokens, H100 runs 3,000 at $0.11/M. GB300 NVL72 is 870% cheaper per token; GB300 NVL72 delivers 1816% more tok/s/chip.
Setting 189 tok/s/user as the target on Qwen 3.5 397B-A17B, GB300 NVL72 produces 37,819 tok/s/chip ($0.017 per million tokens) and H100 produces 2,318 ($0.14). GB300 NVL72 is 726% cheaper per token; GB300 NVL72 delivers 1532% more tok/s/chip. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL7273,544.5H1004,154.2 | GB300 NVL7257,477.6H1002,999.9 | GB300 NVL7237,819.3H1002,317.9 |
| Cost ($/M tok) | GB300 NVL72$0.009H100$0.078 | GB300 NVL72$0.011H100$0.108 | GB300 NVL72$0.017H100$0.140 |
| tok/s/MW | GB300 NVL7234,690,794H1003,032,263 | GB300 NVL7227,112,059H1002,189,711 | GB300 NVL7217,839,270H1001,691,863 |
| Concurrency | GB300 NVL72~273H100~8 | GB300 NVL72~358H100~4 | GB300 NVL72~184H100~4 |