DeepSeek V4 Pro 1.6T — GB300 NVL72 vs MI355X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on DeepSeek V4 Pro 1.6T. 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.
Setting 49 tok/s/user as the target on DeepSeek V4 Pro 1.6T, GB300 NVL72 produces 10183 tok/s/GPU ($0.07 per million tokens) and MI355X produces 1334 ($0.30). GB300 NVL72 is 321% cheaper per token; GB300 NVL72 delivers 663% more tok/s/GPU.
At 83 tok/s/user interactivity on DeepSeek V4 Pro 1.6T, GB300 NVL72 delivers 7828 tok/s/GPU at $0.09 per million tokens; MI355X delivers 687 tok/s/GPU at $0.60. GB300 NVL72 is 535% cheaper per token; GB300 NVL72 delivers 1040% more tok/s/GPU at this point.
GB300 NVL72 posts 4302 tok/s/GPU for $0.17 per million tokens at 117 tok/s/user on DeepSeek V4 Pro 1.6T; MI355X posts 230 tok/s/GPU for $1.79. GB300 NVL72 is 946% cheaper per token; GB300 NVL72 delivers 1768% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:10182.8MI355X:1334.3 | GB300 NVL72:7828.2MI355X:686.9 | GB300 NVL72:4302.4MI355X:230.4 |
| Cost ($/M tok) | GB300 NVL72:$0.072MI355X:$0.305 | GB300 NVL72:$0.094MI355X:$0.599 | GB300 NVL72:$0.171MI355X:$1.790 |
| tok/s/MW | GB300 NVL72:4848964MI355X:503509 | GB300 NVL72:3727700MI355X:259202 | GB300 NVL72:2048747MI355X:86929 |
| Concurrency | GB300 NVL72:~1200MI355X:~25 | GB300 NVL72:~900MI355X:~8 | GB300 NVL72:~365MI355X:~2 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.