I Unlocked My RTX 4090 With a USB Stick. vLLM Throughput Nearly Doubled
NVIDIA ships RTX 20, 30 and 40 series tensor cores at half rate. I unlocked my RTX 4090 with nv-uefi-unlocker and a USB stick, and my vLLM benchmark went from 2,087 to 4,016 tokens/s. Steps taken, before and after benchmarks, and the risks to take note....