Anecdotally, it runs twice as fast and with 20% lower VRAM use on my machines than it did when I first experimented with it in the first week after release. I n

Anecdotally, it runs twice as fast and with 20% lower VRAM use on my machines than it did when I first experimented with it in the first week after release. I no longer need to hyper-optimize memory use (as a layman!) and patch that optimized model into web-frontends that don't come with it to get it to run on a 4gb card, there are flags now that make it "Just Work"™. Down from 90s/512²img -> ~35s. Exited to try it out on some AMD APU's when I've got time to see if that can outperform my dedicated but ancient 4gb card.