How-ToAI ModelsJune 23, 2026

Dual GPU tensor split boosts Qwen3.6-27B to 100+ t/s

Running Qwen3.6-27B at Q8 across an RTX 5090 and 3090 Ti using tensor split-mode yields over 100 tokens per second. The user reports a jump from ~70 t/s after switching from layer split to tensor split.

1 source