2 ms·

VoltaML is making good progress. InvokeAI is also pretty good (but not as optimized/bleeding edge).