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 24–128 tok/s/user interactivity band, at 50 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: GB200 NVL72 runs 9143 tok/s/GPU at $0.07/M tokens, MI355X runs 1099 at $0.37/M. GB200 NVL72 is 458% cheaper per token; GB200 NVL72 delivers 732% more tok/s/GPU.

Setting 76 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 produces 3553 tok/s/GPU ($0.17 per million tokens) and MI355X produces 695 ($0.59). GB200 NVL72 is 243% cheaper per token; GB200 NVL72 delivers 411% more tok/s/GPU.

At 102 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 delivers 451 tok/s/GPU at $1.41 per million tokens; MI355X delivers 482 tok/s/GPU at $0.83. MI355X is 69% cheaper per token; MI355X delivers 7% 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 NVL72:9143.5MI355X:1098.8
GB200 NVL72:3552.7MI355X:695.5
GB200 NVL72:451.0MI355X:482.3
Cost ($/M tok)
GB200 NVL72:$0.067MI355X:$0.375
GB200 NVL72:$0.172MI355X:$0.589
GB200 NVL72:$1.407MI355X:$0.832
tok/s/MW
GB200 NVL72:4889550MI355X:525730
GB200 NVL72:1899865MI355X:332759
GB200 NVL72:241164MI355X:230753
Concurrency
GB200 NVL72:~4301MI355X:~44
GB200 NVL72:~1003MI355X:~19
GB200 NVL72:~98MI355X:~10

Inference Performance

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

Vendor:
Aggregation:
Spec Decoding: