Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on Kimi K3 2.8T. 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 →
At 44 tok/s/user interactivity on Kimi K3 2.8T, GB200 NVL72 delivers 5076 tok/s/chip at $0.10 per million tokens; MI355X delivers 7795 tok/s/chip at $0.05. MI355X is 90% cheaper per token; MI355X delivers 54% more tok/s/chip at this point.
GB200 NVL72 posts 2982 tok/s/chip for $0.17 per million tokens at 80 tok/s/user on Kimi K3 2.8T; MI355X posts 3602 tok/s/chip for $0.12. MI355X is 50% cheaper per token; MI355X delivers 21% more tok/s/chip.
Throughput at 116 tok/s/user on Kimi K3 2.8T: GB200 NVL72 hits 1955 tok/s/chip, MI355X hits 2827. Per-million costs land at $0.26 and $0.15 respectively. MI355X is 79% cheaper per token; MI355X delivers 45% more tok/s/chip. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:5076.3MI355X:7794.5 | GB200 NVL72:2982.5MI355X:3601.9 | GB200 NVL72:1955.0MI355X:2826.8 |
| Cost ($/M tok) | GB200 NVL72:$0.102MI355X:$0.053 | GB200 NVL72:$0.173MI355X:$0.116 | GB200 NVL72:$0.264MI355X:$0.147 |
| tok/s/MW | GB200 NVL72:2714602MI355X:3729449 | GB200 NVL72:1594912MI355X:1723392 | GB200 NVL72:1045450MI355X:1352522 |
| Concurrency | GB200 NVL72:~16MI355X:~15 | GB200 NVL72:~16MI355X:~6 | GB200 NVL72:~13MI355X:~4 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
