DeepSeek V4.1 Flash 552B · Chip comparison
DeepSeek V4.1 Flash 552B — MI325X vs Vera Rubin NVL72
Head-to-head AI inference benchmark comparison of MI325X (AMD CDNA 3) and Vera Rubin NVL72 (NVIDIA Vera Rubin) on DeepSeek V4.1 Flash 552B. 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 →
Setting 255 tok/s/user as the target on DeepSeek V4.1 Flash 552B, MI325X produces 11,985 tok/s/chip ($0.025 per million tokens) and Vera Rubin NVL72 produces 2,657,015 ($0.000). Vera Rubin NVL72 is 99% cheaper per token; Vera Rubin NVL72 delivers 222× the tok/s/chip.
At 330 tok/s/user interactivity on DeepSeek V4.1 Flash 552B, MI325X delivers 8,161 tok/s/chip at $0.037 per million tokens; Vera Rubin NVL72 delivers 2,201,107 tok/s/chip at $0.000. Vera Rubin NVL72 is 99% cheaper per token; Vera Rubin NVL72 delivers 270× the tok/s/chip at this point.
MI325X posts 6,910 tok/s/chip for $0.044 per million tokens at 404 tok/s/user on DeepSeek V4.1 Flash 552B; Vera Rubin NVL72 posts 1,734,610 tok/s/chip for $0.001. Vera Rubin NVL72 is 99% cheaper per token; Vera Rubin NVL72 delivers 251× the tok/s/chip. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MI325X11,985Vera Rubin NVL722,657,015 | MI325X8,161Vera Rubin NVL722,201,107 | MI325X6,910Vera Rubin NVL721,734,610 |
| Cost ($/M tok) | MI325X$0.025Vera Rubin NVL72$0.000 | MI325X$0.037Vera Rubin NVL72$0.000 | MI325X$0.044Vera Rubin NVL72$0.001 |
| tok/s/MW | MI325X7,091,959Vera Rubin NVL72805,156,163 | MI325X4,829,109Vera Rubin NVL72667,002,245 | MI325X4,088,660Vera Rubin NVL72525,639,394 |
| Concurrency | MI325X~6Vera Rubin NVL72~11,551 | MI325X~3Vera Rubin NVL72~11,259 | MI325X~2Vera Rubin NVL72~11,046 |