Qwen3.5 9B
on NVIDIA RTX 5070 Ti Laptop 12GB · Unsloth Studio
Use cases
agenticsummarizationlong-context
Summary
User reports Qwen3.5 9B at around 7.9 tokens/sec on an ASUS ROG Zephyrus G14 with a Ryzen AI 9 HX 370 NPU and 32GB RAM.
Setup is Unsloth Desktop on Windows 11 with the NPU2 build of the model; the RTX 5070 Ti Laptop GPU is present but not used for this figure.
User finds it usable but noticeably slower than cloud models and asks how to get more day-to-day value from NPU inference.
Compare this run
Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.
Qwen3.5 9B at 4-bit by hardware
- 107.5 t/s · one run
- NVIDIA RTX 3080 Ti12 GB100.0 t/s · one run
- 86.5 t/s · one run
- 80.1 t/s · one run
- 68.0 t/s · one run
- 65.5 t/s · one run
Bars are medians on one scale, 0 to 107.5 tokens/s. Press ▶ to watch text arrive at a row's speed.
Everything reported on the NVIDIA RTX 5070 Ti Laptop 12GB →Every Qwen3.5 report →Estimate Qwen3.5 on this card →