Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and MI300X (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.
B300 posts 3084 tok/s/chip for $0.20 per million tokens at 28 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T; MI300X posts 317 tok/s/chip for $0.83. B300 is 308% cheaper per token; B300 delivers 872% more tok/s/chip.
Throughput at 33 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B300 hits 2617 tok/s/chip, MI300X hits 255. Per-million costs land at $0.24 and $1.03 respectively. B300 is 331% cheaper per token; B300 delivers 925% more tok/s/chip.
B300 / MI300X on Kimi K2.5/K2.6/K2.7-Code 1T at 38 tok/s/user: 2278 / 217 tok/s/chip, $0.28 / $1.22 per million tokens. B300 is 341% cheaper per token; B300 delivers 949% more tok/s/chip. (Numbers reflect the default 8k/1k Β· int4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:3083.8MI300X:317.4 | B300:2616.9MI300X:255.2 | B300:2278.5MI300X:217.1 |
| Cost ($/M tok) | B300:$0.204MI300X:$0.831 | B300:$0.240MI300X:$1.034 | B300:$0.276MI300X:$1.215 |
| tok/s/MW | B300:1623061MI300X:228334 | B300:1377293MI300X:183608 | B300:1199206MI300X:156211 |
| Concurrency | B300:~48MI300X:~11 | B300:~35MI300X:~7 | B300:~26MI300X:~5 |
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