Huawei's Kirin 9050 NPU claims offline 30B MoE on smartphones, +140% perf
AIFlow_ML · x · 2026-11-11
Huawei detailed the Kirin 9050 series NPU: a Da Vinci core in its first LogicFolding chip, claiming +140% performance and -66% power vs the previous gen, with a Transformer engine, low-power perception engine, and CANN architecture.
Claimed capabilities:
- Running a 30B MoE model fully offline on a smartphone
- Supporting 6B image models and 7B dense LLMs
- 180% faster text generation with 29% less power
Offline features include eye-tracking page turns, XMage Smart Capture 2.0, and screen-off maps.
More from Infra
- Huawei releases MACaQ: MXFP4 W4A4 PTQ compatible with llama.cpp and vLLM — huawei-csl · 2026-11-11
- Half of 8 recent AI IPOs trade below first-day close; newsletter benchmarks Anthropic IPO and $3.4bn power bet — 0xSammy · 2026-11-11
- Running Qwen2.5-Coder-3B on Android: Jupyter + Pydantic Evals on-device eval pipeline — alex_b3b · 2026-11-11
- Developer ports Stanford's Memorilla memory framework to train Gemma4 locally on Strix Halo — DataPhreak · 2026-11-11
- 22% of Nvidia revenue now books to Singapore, but only ~1% is local buyers — Gregory_C_Allen · 2026-11-11
- Stratechery: Intel vs AMD's diverging earnings and Intel's need for an old-new path — Stratechery · 2026-11-11