StudioHands-on test shows the new Nvidia unlock also works on Chinese 3080 20GB, with ~20% prefill gains
Original title: New Nvidia unlock works on Chinese 3080 20GB too
The author tested Zaggita's new unlock patch on 2x RTX 3080 20GB cards; the patch fixes Nvidia's lockdown of compute speed on consumer cards for specific hardware instructions needed for AI. Prefill speed improved 20.8% to 28.6% (TTFT down 17% to 22%) while decode step rate stayed flat, using vLLM 0.32.0 with tensor parallelism on cyankiwi/Swift-1.5-Qwen3.8-27B-AWQ-INT4. The author also found speeds within 2% between 240W and 320W power limits, and both GPUs hit light thermal throttling at 1750MHz, suggesting roughly 10% more performance with better cooling.
The author ran their own two-GPU test of Zaggita's unlock patch, giving prefill gains, unchanged decode rates, and power and cooling data that Ampere owners can directly compare against.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.