Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on DeepSeek V4.1 Flash 552B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics β 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 β
At 189 tok/s/user interactivity on DeepSeek V4.1 Flash 552B, GB300 NVL72 delivers 652718 tok/s/chip at $0.00 per million tokens; MI355X delivers 66875 tok/s/chip at $0.01. GB300 NVL72 is 534% cheaper per token; GB300 NVL72 delivers 876% more tok/s/chip at this point.
GB300 NVL72 posts 19067 tok/s/chip for $0.03 per million tokens at 335 tok/s/user on DeepSeek V4.1 Flash 552B; MI355X posts 23178 tok/s/chip for $0.02. MI355X is 87% cheaper per token; MI355X delivers 22% more tok/s/chip.
Throughput at 482 tok/s/user on DeepSeek V4.1 Flash 552B: GB300 NVL72 hits 9938 tok/s/chip, MI355X hits 14114. Per-million costs land at $0.06 and $0.03 respectively. MI355X is 119% cheaper per token; MI355X delivers 42% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:652717.8MI355X:66875.5 | GB300 NVL72:19067.0MI355X:23177.8 | GB300 NVL72:9937.7MI355X:14114.2 |
| Cost ($/M tok) | GB300 NVL72:$0.001MI355X:$0.006 | GB300 NVL72:$0.034MI355X:$0.018 | GB300 NVL72:$0.065MI355X:$0.030 |
| tok/s/MW | GB300 NVL72:307885750MI355X:31997829 | GB300 NVL72:8993879MI355X:11089859 | GB300 NVL72:4687617MI355X:6753223 |
| Concurrency | GB300 NVL72:~3254MI355X:~32 | GB300 NVL72:~16MI355X:~6 | GB300 NVL72:~4MI355X:~1 |
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.
