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 β same interactions as the main inference chart.
At 28 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, B300 delivers 3084 tok/s/chip at $0.20 per million tokens; MI325X delivers 413 tok/s/chip at $0.74. B300 is 264% cheaper per token; B300 delivers 647% more tok/s/chip at this point.
B300 posts 2537 tok/s/chip for $0.25 per million tokens at 34 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T; MI325X posts 316 tok/s/chip for $0.97. B300 is 291% cheaper per token; B300 delivers 703% more tok/s/chip.
Throughput at 40 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B300 hits 2181 tok/s/chip, MI325X hits 247. Per-million costs land at $0.29 and $1.24 respectively. B300 is 329% cheaper per token; B300 delivers 782% more tok/s/chip. (Numbers reflect the default 8k/1k Β· int4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:3083.8MI325X:412.6 | B300:2537.1MI325X:315.9 | B300:2180.9MI325X:247.3 |
| Cost ($/M tok) | B300:$0.204MI325X:$0.741 | B300:$0.247MI325X:$0.967 | B300:$0.288MI325X:$1.235 |
| tok/s/MW | B300:1623061MI325X:244132 | B300:1335311MI325X:186945 | B300:1147838MI325X:146352 |
| Concurrency | B300:~48MI325X:~14 | B300:~33MI325X:~9 | B300:~24MI325X:~6 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom β’ Drag to pan β’ Double-click to reset β’ Click a point to pin tooltip
Kimi K2.5/2.6/2.7-Code 1T 8K / 1K Total Tokens per $1 TCO vs. Interactivity