3 ms·

For cutting-edge LLM work, probably? I mean, I run mine on older hardware than that, but I'm a total hobbyist...