← sources

What is your experience with inference engines under Windows?

source: reddit (reddit:LocalLLaMA)published 2026-11-16 17:21ingested 2026-11-17 02:41

retained text

As the title says: what is your experience using all the new inference engines under windows?  
  
Currently I'm using a laptop with this hardware:

CPU: Intel 275HX  
RAM: 2x32gb of ddr5 5600  
GPU: RXT 5090 (24gb VRAM)  
SSD: Samsung 990 pro 2tb + SK Hynix PCB01 1tb

I'm running Windows 11 Home edition with a pretty clean installation.

I've tried quite a lot of the new vibecoded inference engines but I always end up going back to Strata for Qwen 3.8 Flash Next and to beellama.cpp for everthing else.

I've tried nInfer, basalt, rusting strata, exLlamaV3 and pretty much any fork of llama.cpp and they all end up being way slower than what I'm already using. I've lost a little bit of time on each and also tried to use Claude or Codex to optimize the configurations but there is just no way to get decent performance from engines that on paper should be at least as fast as what I'm currently using.

I'm doing something horribly wrong or they work well only on a linux system?

document fetched 2026-11-17 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xxv9f2/what_is_your_experience_with_inference_engines