Kimi K2.5/K2.6/K2.7-Code 1T · GPU comparison

Kimi K2.5/K2.6/K2.7-Code 1T — GB200 NVL72 vs MI355X

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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.

Near the low end of the 24–113 tok/s/user interactivity band, at 46 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: GB200 NVL72 runs 10065 tok/s/GPU at $0.06/M tokens, MI355X runs 726 at $0.56/M. GB200 NVL72 is 823% cheaper per token; GB200 NVL72 delivers 1286% more tok/s/GPU.

Setting 68 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 produces 5984 tok/s/GPU ($0.10 per million tokens) and MI355X produces 554 ($0.74). GB200 NVL72 is 625% cheaper per token; GB200 NVL72 delivers 980% more tok/s/GPU.

At 91 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 delivers 600 tok/s/GPU at $1.05 per million tokens; MI355X delivers 419 tok/s/GPU at $0.98. MI355X is 7% cheaper per token; GB200 NVL72 delivers 43% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)

View performance-per-dollar view →

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/gpu)
GB200 NVL72:10064.8MI355X:726.2
GB200 NVL72:5983.5MI355X:553.9
GB200 NVL72:599.6MI355X:418.8
Cost ($/M tok)
GB200 NVL72:$0.061MI355X:$0.565
GB200 NVL72:$0.102MI355X:$0.741
GB200 NVL72:$1.047MI355X:$0.982
tok/s/MW
GB200 NVL72:4792752MI355X:274039
GB200 NVL72:2849292MI355X:209003
GB200 NVL72:285506MI355X:158047
Concurrency
GB200 NVL72:~4301MI355X:~35
GB200 NVL72:~2093MI355X:~17
GB200 NVL72:~180MI355X:~9

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.

Vendor:
Aggregation:
Spec Decoding: