RTX 3060 LHR unlock: BF16 +85.5% in benchmarks but no gain in ComfyUI video gen
JinnTsui · reddit · 2026-11-06
A Redditor unlocked the LHR limit on an RTX 3060 12GB (GA106) via Windows-GPU-UNLOCK, measuring BF16 +85.5% and FP16 +102.2% in PyTorch matmul benchmarks — yet saw no noticeable speedup in a MiniMax H3 INT8 video-generation workflow. Key takeaway: more Tensor Core throughput doesn't automatically speed up every AI workload.
More from Infra
- USM turns idle NVIDIA VRAM into Linux swap, ~9 µs vs 55 µs on NVMe — kha333n · 2026-11-06
- Opening a KTV next to Ulanqab's AI data centers is the new 'sure-fire' business, jokes Chinese AI circle — basedjensen · 2026-11-06
- Musk claims SoftBank is defaulting on its turbine orders for AI datacenters — SumitGup · 2026-11-06
- Thinking Machines' Inkling 2 Small trained on 12.55T tokens, with RL consuming over a quarter of pretraining compute — YouJiacheng · 2026-11-06
- SpaceX plans ~6 GW Arizona datacenter across 9 buildings, no industrial water — elonmusk · 2026-11-06
- DeepSeek-V4.1-Flash (552B) runs at 38.2 tps on a single AMD Ryzen AI Max+ 395 mini PC — solyarisoftware · 2026-11-06