GenieX brings any GGUF model to Snapdragon's Hexagon NPU, with real benchmarks
SkyFeistyLlama8 · reddit · 2026-10-28
A detailed Reddit writeup tests Qualcomm's GenieX (built on work from the acquired Nexa SDK team), which combines llama.cpp's Hexagon NPU and Adreno OpenCL backends with signed drivers to run any GGUF model on Snapdragon X/X2 Plus/Elite chips across Windows, Linux and Android, plus an OpenAI-compatible llama-server.
Benchmarks (ThinkPad T14s, X1E-78-100, 64GB RAM):
- NPU prefill is 2x faster than CPU (1616 vs 893 tps) at 12W vs 35-60W — roughly 10x more efficient
- NPU decode is only 32 tps, half of CPU, likely memory-bandwidth limited
- Gemma4-26B MoE runs 5 t/s on NPU at just 10W silently; 31B dense gets only 1 t/s on NPU vs 5 t/s on throttling CPU
Takeaway: the NPU shines for prefill and low-power inference, while decode remains the weak spot.
More from Infra
- Mixed NVIDIA+AMD multi-GPU LLM setup benchmarks slower than a single card — LMasterGame · 2026-10-28
- 8 open-source repos to self-host a lean AI agent stack instead of stacking $20/month SaaS — SaiTarunRonanki · 2026-10-28
- Corning Optical Q3: sales up 44% to $2.38B, Gen AI revenue doubles y/y — demian_ai · 2026-10-28
- AI-powered formal verification finds real Redis/Valkey bugs, incl. a 23-char freeze — LingmingZhang · 2026-10-28
- 112 TPS on a single RTX 3090: LlamAmpere v0.6 pushes 27B models with 400K context — Brief-Tap-6616 · 2026-10-28
- Your coding agent burns up to 12k tokens before you type: 30 harnesses span 50x — yb2698 · 2026-10-28