Qwen 3.5 397B-A17B · Chip comparison
Qwen 3.5 397B-A17B — GB300 NVL72 vs MI355X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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 / MI355X on Qwen 3.5 397B-A17B at 129 tok/s/user: 64,352 / 31,756 tok/s/chip, $0.010 / $0.013 per million tokens. GB300 NVL72 is 32% cheaper per token; GB300 NVL72 delivers 103% more tok/s/chip.
Around the middle of the 68–314 tok/s/user interactivity band, at 191 tok/s/user on Qwen 3.5 397B-A17B: GB300 NVL72 runs 37,281 tok/s/chip at $0.017/M tokens, MI355X runs 17,266 at $0.024/M. GB300 NVL72 is 40% cheaper per token; GB300 NVL72 delivers 116% more tok/s/chip.
Setting 253 tok/s/user as the target on Qwen 3.5 397B-A17B, GB300 NVL72 produces 28,087 tok/s/chip ($0.023 per million tokens) and MI355X produces 6,014 ($0.069). GB300 NVL72 is 203% cheaper per token; GB300 NVL72 delivers 367% more tok/s/chip. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL7264,351.6MI355X31,756.0 | GB300 NVL7237,281.4MI355X17,266.3 | GB300 NVL7228,087.3MI355X6,013.8 |
| Cost ($/M tok) | GB300 NVL72$0.010MI355X$0.013 | GB300 NVL72$0.017MI355X$0.024 | GB300 NVL72$0.023MI355X$0.069 |
| tok/s/MW | GB300 NVL7230,354,517MI355X15,194,245 | GB300 NVL7217,585,556MI355X8,261,398 | GB300 NVL7213,248,711MI355X2,877,427 |
| Concurrency | GB300 NVL72~387MI355X~24 | GB300 NVL72~181MI355X~12 | GB300 NVL72~131MI355X~4 |