KohakUnlock removes RTX 20-series FP16→FP32 lock: 2080 Ti jumps from 59.9 to 111 TFLOPS
CeFurkan · reddit · 2026-10-25
The open-source KohakUnlock project unlocks the capped FP16→FP32 mixed-precision throughput on RTX 2000-series GPUs. README benchmarks (screenshot): 2080 Ti FP16→FP32 mmapeak rises from 59.9 to 111.0 TFLOPS (+85%), while the FP16→FP16 control is unchanged (119.5 vs 118.5 TFLOPS), confirming an unlock rather than a driver quirk; best-MAMF median across a 125-shape sweep climbs from 57.71 to 88.24 TFLOPS; a single 3090 hits 122.76 TFLOPS (up from 73.03) in PyTorch GEMM, with power limits at 370W stock / 360W unlocked.
More from Infra
- 'For rent: GB300s, never used' — two-sentence horror about idle Blackwell GPUs — sloppenheimer · 2026-10-25
- DeepSeek-V4.1-Flash EXL3-3bpw quant runs on 2x RTX PRO 6000 — No-Paper-557 · 2026-10-25
- Homelabber finishes 8x RTX 3090 build: 192GB VRAM + 256GB RAM for local LLMs — generativist · 2026-10-25
- Beyond funding, thermodynamics is the real hard limit on AI progress — AryHHAry · 2026-10-25
- Aggregated 18 ComfyUI GPU-unlock benchmarks show BF16 gains dwarf INT8 — Acceptable_Put9983 · 2026-10-25
- Blogger revisits Kimi K3 hype, speculates China may release some B300 GPUs — lipeng0820 · 2026-10-25