Qwen 3.8 Flash Next 176B-A6B · Chip comparison
Qwen 3.8 Flash Next 176B-A6B — TPU7x vs Vera Rubin NVL72
Head-to-head AI inference benchmark comparison of TPU7x (Google Ironwood) and Vera Rubin NVL72 (NVIDIA Vera Rubin) on Qwen 3.8 Flash Next 176B-A6B. 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.
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 →
Vera Rubin NVL72: 210,418 tok/s/chip, $0.005 per million tokens at 246 tok/s/user on Qwen 3.8 Flash Next 176B-A6B. TPU7x is unmeasured here.
At 401 tok/s/user on Qwen 3.8 Flash Next 176B-A6B, Vera Rubin NVL72 delivers 162,255 tok/s/chip at $0.006 per million tokens; TPU7x hasn't been benchmarked at this target.
Vera Rubin NVL72 hits 37,355 tok/s/chip for $0.027 per million tokens at 555 tok/s/user on Qwen 3.8 Flash Next 176B-A6B. No TPU7x data at this operating point. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | TPU7x—Vera Rubin NVL72210,418 | TPU7x—Vera Rubin NVL72162,255 | TPU7x—Vera Rubin NVL7237,355 |
| Cost ($/M tok) | TPU7x—Vera Rubin NVL72$0.005 | TPU7x—Vera Rubin NVL72$0.006 | TPU7x—Vera Rubin NVL72$0.027 |
| tok/s/MW | TPU7x—Vera Rubin NVL7263,763,060 | TPU7x—Vera Rubin NVL7249,168,280 | TPU7x—Vera Rubin NVL7211,319,609 |
| Concurrency | TPU7x—Vera Rubin NVL72~129 | TPU7x—Vera Rubin NVL72~30 | TPU7x—Vera Rubin NVL72~5 |