DeepSeek V4.1 Flash 552B · Chip comparison

DeepSeek V4.1 Flash 552B — MI300X vs Vera Rubin NVL72

Head-to-head AI inference benchmark comparison of MI300X (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 →

At 190 tok/s/user interactivity on DeepSeek V4.1 Flash 552B, MI300X delivers 2,310 tok/s/chip at $0.11 per million tokens; Vera Rubin NVL72 delivers 2,953,819 tok/s/chip at $0.000. Vera Rubin NVL72 is 99.7% cheaper per token; Vera Rubin NVL72 delivers 1279× the tok/s/chip at this point.

MI300X posts 2,237 tok/s/chip for $0.12 per million tokens at 199 tok/s/user on DeepSeek V4.1 Flash 552B; Vera Rubin NVL72 posts 2,914,990 tok/s/chip for $0.000. Vera Rubin NVL72 is 99.7% cheaper per token; Vera Rubin NVL72 delivers 1303× the tok/s/chip.

Throughput at 208 tok/s/user on DeepSeek V4.1 Flash 552B: MI300X hits 2,194 tok/s/chip, Vera Rubin NVL72 hits 2,875,770. Costs per million tokens land at $0.12 and $0.000 respectively. Vera Rubin NVL72 is 99.7% cheaper per token; Vera Rubin NVL72 delivers 1311× 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.)

View performance per dollar →

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)
MI300X2,310Vera Rubin NVL722,953,819
MI300X2,237Vera Rubin NVL722,914,990
MI300X2,194Vera Rubin NVL722,875,770
Cost ($/M tok)
MI300X$0.11Vera Rubin NVL72$0.000
MI300X$0.12Vera Rubin NVL72$0.000
MI300X$0.12Vera Rubin NVL72$0.000
tok/s/MW
MI300X1,662,039Vera Rubin NVL72895,096,579
MI300X1,609,518Vera Rubin NVL72883,330,337
MI300X1,578,219Vera Rubin NVL72871,445,514
Concurrency
MI300X~2Vera Rubin NVL72~15,084
MI300X~2Vera Rubin NVL72~14,398
MI300X~1Vera Rubin NVL72~13,746