← sources

Model recommendations for cybersec work on 5090+5080 multi gpu setup + 32GB sys RAM

source: reddit (reddit:LocalLLaMA)published 2026-11-03 11:44ingested 2026-11-04 02:41

retained text

As title says. 

Just wondering what you all would recommend, as well as, inference engine recommendations. 

Currently I’m using megacapybara for a qwen 3.8 27B variant. But was wondering if there was any other inference engines or anything better for running larger models.

Did see a network chuck recommendation for a gpu setup that allows DeepSeek 4.1 flash to run on a single 5090 but idk what the decode t/s would be on that. 

Anything other info is much appreciated! 

document fetched 2026-11-04 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1xmqt4u/model_recommendations_for_cybersec_work_on