Kimi K2.5/K2.6/K2.7-Code 1T — B300 vs MI325X
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on Kimi K2.5/K2.6/K2.7-Code 1T. 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.
At 33 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, B300 delivers 722 tok/s/GPU at $0.91 per million tokens; MI325X delivers 127 tok/s/GPU at $2.79. B300 is 207% cheaper per token; B300 delivers 467% more tok/s/GPU at this point.
B300 posts 603 tok/s/GPU for $1.08 per million tokens at 39 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T; MI325X posts 94 tok/s/GPU for $3.80. B300 is 252% cheaper per token; B300 delivers 539% more tok/s/GPU.
Throughput at 45 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B300 hits 535 tok/s/GPU, MI325X hits 68. Per-million costs land at $1.22 and $5.25 respectively. B300 is 329% cheaper per token; B300 delivers 687% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:722.3MI325X:127.3 | B300:602.6MI325X:94.2 | B300:535.2MI325X:68.0 |
| Cost ($/M tok) | B300:$0.910MI325X:$2.789 | B300:$1.078MI325X:$3.800 | B300:$1.224MI325X:$5.251 |
| tok/s/MW | B300:332852MI325X:58415 | B300:277716MI325X:43231 | B300:246648MI325X:31189 |
| Concurrency | B300:~47MI325X:~16 | B300:~32MI325X:~10 | B300:~22MI325X:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.