← sources

ISTA Qwen3.8 27b on 3060 12gb

source: reddit (reddit:LocalLLaMA)published 2026-11-15 14:44ingested 2026-11-16 02:38

retained text

Hey guys 

I am a seasoned local llm user and I ain't quite ram rich like only 16gb ddr4 🥀

So I can't run qwen3.8 flash next on my 3060 

This guy felt the best for me yet 

Have you guys tried it 

I have ran iq3\_xss quant non mtp with turboquant at 65k context completely on my 3060 though I did offload mmproj to cpu ram 

Have guys tried this it really felt like a blessing to my life using the peak of local ai on my 3060 

It seriously packs the punch 👊 I was able to perform various of my agent task complete on device not like simple but the ones which require actual reasonings 

Here is my cmd if you want to see

llama-tq serve -m "E:\\MODELS\\ISTA-DASLab\\Qwen3.8-27B-GSQ-RCO-GGUF\\Qwen3.8-27B-GSQ-RCO-IQ3\_XXS.gguf" --mmproj "E:\\MODELS\\ISTA-DASLab\\Qwen3.8-27B-GSQ-RCO-GGUF\\mmproj-Qwen3.8-27B-BF16.gguf" --no-mmproj-offload -ngl 99 -c 65536 -ctk turbo4 -ctv turbo3 -ub 256 -b 2048 -t 3 -tb 3 --reasoning-budget 256 --reasoning-preserve --alias Qwen --port 8080 -lm none

Sources: for my recipe

ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF

https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/discussions

Turboquant

https://github.com/TheTom/llama-cpp-turboquant

Pi coding agent

https://pi.dev/

I would like know the communities views and any suggestions would be welcome 🤗

Thankyou for your time 

document fetched 2026-11-16 00:37 · canonical https://reddit.com/r/LocalLLaMA/comments/1xwx7uq/ista_qwen38_27b_on_3060_12gb