DeepSeek V4.1 Flash 552B β GB300 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) 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
GB300 NVL72 / MI325X on DeepSeek V4.1 Flash 552B at 85 tok/s/user: 1094347 / 13589 tok/s/chip, $0.00 / $0.02 per million tokens. GB300 NVL72 is 3735% cheaper per token; GB300 NVL72 delivers 7953% more tok/s/chip.
Around the middle of the 44β209 tok/s/user interactivity band, at 127 tok/s/user on DeepSeek V4.1 Flash 552B: GB300 NVL72 runs 1032799 tok/s/chip at $0.00/M tokens, MI325X runs 6724 at $0.05/M. GB300 NVL72 is 7214% cheaper per token; GB300 NVL72 delivers 15259% more tok/s/chip.
Setting 168 tok/s/user as the target on DeepSeek V4.1 Flash 552B, GB300 NVL72 produces 975015 tok/s/chip ($0.00 per million tokens) and MI325X produces 4218 ($0.07). GB300 NVL72 is 10907% cheaper per token; GB300 NVL72 delivers 23016% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:1094347.1MI325X:13588.9 | GB300 NVL72:1032798.5MI325X:6724.3 | GB300 NVL72:975014.5MI325X:4218.0 |
| Cost ($/M tok) | GB300 NVL72:$0.001MI325X:$0.022 | GB300 NVL72:$0.001MI325X:$0.045 | GB300 NVL72:$0.001MI325X:$0.072 |
| tok/s/MW | GB300 NVL72:516201439MI325X:8040779 | GB300 NVL72:487169116MI325X:3978860 | GB300 NVL72:459912514MI325X:2495857 |
| Concurrency | GB300 NVL72:~7480MI325X:~5 | GB300 NVL72:~7480MI325X:~1 | GB300 NVL72:~8883MI325X:~1 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.