Llama 3.3 70B — B200 vs MI300X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI300X (AMD CDNA 3) on Llama 3.3 70B. 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.
Near the low end of the 34–112 tok/s/user interactivity band, at 53 tok/s/user on Llama 3.3 70B: B200 runs 6496 tok/s/GPU at $0.08/M tokens, MI300X runs 1185 at $0.26/M. B200 is 208% cheaper per token; B200 delivers 448% more tok/s/GPU.
Setting 73 tok/s/user as the target on Llama 3.3 70B, B200 produces 4657 tok/s/GPU ($0.12 per million tokens) and MI300X produces 660 ($0.48). B200 is 312% cheaper per token; B200 delivers 606% more tok/s/GPU.
At 93 tok/s/user interactivity on Llama 3.3 70B, B200 delivers 3478 tok/s/GPU at $0.15 per million tokens; MI300X delivers 373 tok/s/GPU at $0.83. B200 is 444% cheaper per token; B200 delivers 832% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:6496.2MI300X:1185.2 | B200:4657.0MI300X:660.0 | B200:3477.7MI300X:372.9 |
| Cost ($/M tok) | B200:$0.083MI300X:$0.257 | B200:$0.116MI300X:$0.479 | B200:$0.153MI300X:$0.832 |
| tok/s/MW | B200:2993645MI300X:662142 | B200:2146065MI300X:368695 | B200:1602627MI300X:208350 |
| Concurrency | B200:~128MI300X:~50 | B200:~128MI300X:~32 | B200:~86MI300X:~17 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.