Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on DeepSeek R1. 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.
Throughput at 37 tok/s/user on DeepSeek R1: GB200 NVL72 hits 7073 tok/s/chip, MI325X hits 646. Per-million costs land at $0.07 and $0.47 respectively. GB200 NVL72 is 547% cheaper per token; GB200 NVL72 delivers 994% more tok/s/chip.
GB200 NVL72 / MI325X on DeepSeek R1 at 43 tok/s/user: 6988 / 500 tok/s/chip, $0.07 / $0.61 per million tokens. GB200 NVL72 is 726% cheaper per token; GB200 NVL72 delivers 1297% more tok/s/chip.
Toward the upper edge of the 32–53 tok/s/user interactivity band, at 48 tok/s/user on DeepSeek R1: GB200 NVL72 runs 6861 tok/s/chip at $0.08/M tokens, MI325X runs 371 at $0.82/M. GB200 NVL72 is 994% cheaper per token; GB200 NVL72 delivers 1750% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:7072.6MI325X:646.3 | GB200 NVL72:6988.1MI325X:500.1 | GB200 NVL72:6861.5MI325X:370.9 |
| Cost ($/M tok) | GB200 NVL72:$0.073MI325X:$0.473 | GB200 NVL72:$0.074MI325X:$0.611 | GB200 NVL72:$0.075MI325X:$0.824 |
| tok/s/MW | GB200 NVL72:3782124MI325X:382442 | GB200 NVL72:3736945MI325X:295910 | GB200 NVL72:3669226MI325X:219492 |
| Concurrency | GB200 NVL72:~2484MI325X:~17 | GB200 NVL72:~4883MI325X:~11 | GB200 NVL72:~6142MI325X:~8 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip
DeepSeek R1 0528 671B 8K / 1K Total Tokens per $1 TCO vs. Interactivity