Head-to-head AI inference benchmark comparison of MI325X (AMD CDNA 3) and TPU7x (Google Ironwood) on Qwen 3.5 397B-A17B. 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.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX β
MI325X / TPU7x on Qwen 3.5 397B-A17B at 141 tok/s/user: 3631 / 37928 tok/s/chip, $0.08 / $0.01 per million tokens. TPU7x is 1016% cheaper per token; TPU7x delivers 945% more tok/s/chip.
Around the middle of the 2β558 tok/s/user interactivity band, at 280 tok/s/user on Qwen 3.5 397B-A17B: MI325X runs 3631 tok/s/chip at $0.08/M tokens, TPU7x runs 26053 at $0.01/M. TPU7x is 666% cheaper per token; TPU7x delivers 618% more tok/s/chip.
Setting 419 tok/s/user as the target on Qwen 3.5 397B-A17B, MI325X produces 3631 tok/s/chip ($0.08 per million tokens) and TPU7x produces 18182 ($0.02). TPU7x is 435% cheaper per token; TPU7x delivers 401% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp8 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MI325X:3630.9TPU7x:37928.5 | MI325X:3630.9TPU7x:26052.9 | MI325X:3630.9TPU7x:18182.1 |
| Cost ($/M tok) | MI325X:$0.084TPU7x:$0.008 | MI325X:$0.084TPU7x:$0.011 | MI325X:$0.084TPU7x:$0.016 |
| tok/s/MW | MI325X:2148478TPU7x:31423770 | MI325X:2148478TPU7x:21584818 | MI325X:2148478TPU7x:15063841 |
| Concurrency | MI325X:~8TPU7x:~26 | MI325X:~8TPU7x:~16 | MI325X:~8TPU7x:~11 |
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.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
Shift+Scroll to zoom β’ Drag to pan β’ Double-click to reset β’ Click a point to pin tooltip
Qwen3.5 397B 8K / 1K Total Tokens per $1 TCO vs. Interactivity