← sources

Quick One Shot Pelican RTX 5090 Test - Qwen 3.8 27B. Q4_K_M vs Q5_K_M Quants. Both Ran @ 110-120tokens/sec in OpenCode 2+LM Studio.

source: reddit (reddit:LocalLLaMA)published 2026-11-09 01:33ingested 2026-11-10 02:42

retained text

Test - A pelican riding a bicycle.  
GPU - RTX 5090 (31GB VRAM Usage, Model+KV cache)  
Speed - 110-120tok/sec  
Model - Qwen 3.8 27B Unsloth, GGUF  
Quants/Context - Q4\_K\_M/120K vs. Q5\_K\_M/90K  
Software - OpenCode 2+LM Studio(MTP On, Thinking Medium, All Else Default)

document fetched 2026-11-10 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1xrexe0/quick_one_shot_pelican_rtx_5090_test_qwen_38_27b