Originally Posted by Vantavia
Think of it as frame latency for the CPU. AFAIK it's a measure of responsiveness/consistency. It may turn out that frame latency and DPC latency are closely related, but I can't graph DPC over time to test that theory.
OK, I follow you, thanks for the analogy. I wonder if they could be inversely related, at least on some systems? I've noticed (again, a small but consistent) frametime improvement with HT enabled. Now HT definitely increases CPU latency, but it allows more processes to complete in a given time. Increased throughput, yes, but increased latency for individual processes comes along with it. The effect for me of enabling HT is slightly smoother game play despite the fact that it's not what I expected to see.
I also get a small improvement by changing Win32PrioritySeparation from the default 26 to a value of 16. 26 gives a short, variable quantum and a 3x foreground boost. 16 gives a long, variable quantum and a 3x foreground boost. The longer quantum means that more foreground processes will finish first time around, but it also means that others must wait longer to get any CPU time. A short one means that every process gets some CPU time asap, but fewer finish on the first go and the CPU spends more time switching from one partially-completed process to the next. A short quantum strikes me as something that ought improve responsiveness and reduce CPU latency, but for my purposes, at least, it doesn't help with gaming.
I am sure that *all* of this depends on what you do with your PC. Something that improves video transcoding is not necessarily something that will help with factoring primes or with modeling hurricanes or with gaming, etc, and one system might see different benefits or problems than another system using the same tweak. So I say, tweak away, but forget benchmarks and other synthetic tests, and use a measure that actually reflects what you do with your system.
Wow, that was a wall of text, lol