Qwen 3.5 397B-A17B · Chip comparison
Qwen 3.5 397B-A17B — B300 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) 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 →
B300 / GB300 NVL72 on Qwen 3.5 397B-A17B at 148 tok/s/user: 34,184 / 57,478 tok/s/chip, $0.018 / $0.011 per million tokens. GB300 NVL72 is 65% cheaper per token; GB300 NVL72 delivers 68% more tok/s/chip.
Around the middle of the 68–389 tok/s/user interactivity band, at 228 tok/s/user on Qwen 3.5 397B-A17B: B300 runs 18,270 tok/s/chip at $0.034/M tokens, GB300 NVL72 runs 31,170 at $0.021/M. GB300 NVL72 is 67% cheaper per token; GB300 NVL72 delivers 71% more tok/s/chip.
Setting 309 tok/s/user as the target on Qwen 3.5 397B-A17B, B300 produces 8,886 tok/s/chip ($0.071 per million tokens) and GB300 NVL72 produces 23,234 ($0.028). GB300 NVL72 is 156% cheaper per token; GB300 NVL72 delivers 161% more tok/s/chip. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
GB300 NVL72 vs B300: full specs & benchmarks across models →
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B30034,183.6GB300 NVL7257,477.6 | B30018,270.3GB300 NVL7231,170.0 | B3008,886.2GB300 NVL7223,233.6 |
| Cost ($/M tok) | B300$0.018GB300 NVL72$0.011 | B300$0.034GB300 NVL72$0.021 | B300$0.071GB300 NVL72$0.028 |
| tok/s/MW | B30017,991,377GB300 NVL7227,112,059 | B3009,615,934GB300 NVL7214,702,811 | B3004,676,950GB300 NVL7210,959,228 |
| Concurrency | B300~26GB300 NVL72~358 | B300~26GB300 NVL72~148 | B300~12GB300 NVL72~105 |