Community
Reddit user reports disappointing results from Strata inference engine on two R9700 GPUs
A LocalLLaMA user found Strata running Qwen Next Flash on 2x R9700 hung on long tasks unless experts fit in VRAM, and a 3-bit quant scored 93.9% on HumanEval+ versus 95.1% for Qwen3.8-27B Q8 on llama.cpp while using nearly twice the tokens.
Every outlet that ran it 1 article
r/LocalLLaMACommunityLead
Are bespoke inference engines overhyped? My Strata experience on 2× R9700