Kimi K3 2.8T — GB200 NVL72 vs MI355X Performance per Dollar
Cost per million tokens of GB200 NVL72 (NVIDIA Blackwell) versus MI355X (AMD CDNA 4) on Kimi K3 2.8T. Owning-hyperscaler TCO normalized by total tokens — performance per dollar across LLM workloads. Pick the more cost-efficient SKU at every target interactivity level. 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 →
MI355X edges GB200 NVL72 at 36 tok/s/user on Kimi K3 2.8T — $0.06 per million tokens versus $0.09, a 56% cost-per-token gap.
Push Kimi K3 2.8T to 64 tok/s/user and GB200 NVL72 lands at $0.14 per million tokens against MI355X's $0.12 — MI355X pulls ahead by 16%.
GB200 NVL72: $0.20 per million tokens. MI355X: $0.22. Both at 92 tok/s/user on Kimi K3 2.8T, with GB200 NVL72 10% cheaper. (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.)
Chip pricing (owning hyperscaler): GB200 NVL72 $1.86/chip/hr · MI355X $1.50/chip/hr. Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Dollar per Million Tokens | GB200 NVL72:$0.092MI355X:$0.059 | GB200 NVL72:$0.141MI355X:$0.121 | GB200 NVL72:$0.203MI355X:$0.224 |
| Concurrency | GB200 NVL72:~21MI355X:~21 | GB200 NVL72:~16MI355X:~5 | GB200 NVL72:~16MI355X:~4 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
No data available
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.