Can I install Qwen3.6 27b on my computer?
Posted by drazyan22@reddit | LocalLLaMA | View on Reddit | 21 comments
Here is my computer. I want to use Qwen3.6 27b for coding but my GPU just have 16gb Vram. Can I install it or not? or I need to sell GPU and buy new one with 24gb Vram?
LocalLLaMA-ModTeam@reddit
Rule 1 - Search before asking.
sagiroth@reddit
Too much quant to run on 16gb vram. Better off 35A3B. Otherwise vram 24 gpu
drazyan22@reddit (OP)
Thanks for your reply. I think I'll sell my GPU and buy new one
LagOps91@reddit
Better run the 124b model actually since it's an MoE and you can make use of your 64gb ram (if I read that correctly). Q4 should fit.
Finanzamt_kommt@reddit
Do you still have an old gpu lying around? Because with that you get a small slowdown compared better gpus but in vram is always faster than if you spill over.
drazyan22@reddit (OP)
Unfortunately, I don’t have any spare GPU. I might have to stick with a smaller model or upgrade later.
sagiroth@reddit
Terrible time to buy one tbh, I would advise on used 3090 if u can find one at decent price over new
Academic-Map268@reddit
You can fit Q3_K_M in your VRAM and offload the KV cache. It should be smarter than the 35B since dense models are less sensitive to quantization.
Long_comment_san@reddit
You need to download more ram.
Use 35b a3b, it's basically designed for system like ours (I have 64+12)
Lost-Health-8675@reddit
But with ram prices find a cracked one, so you get it for free
drazyan22@reddit (OP)
thank you
KURD_1_STAN@reddit
U can run q3 km or maybe just maybe run q3 kxl with lower context. But wait for gguf chart to see if they are worth it.
drazyan22@reddit (OP)
Thank for you advice
KURD_1_STAN@reddit
If they made qwen3.6 122b then u should try that, it is perfect for ur system. Cause this is really not the time to buy any xomputer hardware, or it it is the best time cause it can get 10 times worse, who knows.
Technical-Earth-3254@reddit
You can, but you should go for the 35B with the 16GB 5080.
drazyan22@reddit (OP)
thank you. I will try it
Technical-Earth-3254@reddit
If you don't wanna go through all the struggle of setting up an environment, just download unsloth or lm studio.
jacek2023@reddit
Yes I was able to run it on 5070
drazyan22@reddit (OP)
thanks a lot
jacek2023@reddit
start from this quant https://www.reddit.com/r/LocalLLaMA/comments/1sstxhk/coding_with_qwen3627budq2_k_xlgguf/
Skyline34rGt@reddit
Better and faster option probably be higher Qwen3.6 35b-a4b quant.
With 27b you will need very low quant (remember of lonx context needed vram)