Originally Posted by Farih
But do you think spikes to 500+us is fine ?
Try a 24bit 96khz FLAC file, i reckon you get audio dropouts then.
Takes really major issues to cause audio dropouts while doing nothing but playing an audio file.
I can probably find or setup a system with persistent 5000+ microsecond DPC latency that won't produce a single dropout, crack, or pop while playing a 24-bit, 96kHz FLAC file...but which will have very obvious audio problems with increasing numbers of streams, especially with other loads going on.
Originally Posted by Sele1981
By latency problems, I need very low latency times because I play racing simulations and need quick reaction times.
DPC execution times don't imply much about input latency.
Originally Posted by JackCY
Still Nvidia driver shows as the highest DPC, it's not bad absolute value but it still shows.
71 microseconds is an order of magnitude below the point where most anyone will be able to detect related issues in virtually anything.
Originally Posted by Dargonplay
HPET is a performance and latency hog if you're not running a bank of servers that need to have their timings synchronized.
TSC is another timer system ingrained in Windows and our BIOSes, it is in fact x100 times faster than HPET in certain scenarios (when multiple interrupts are being handled at the same time in chaos), it is a more "real time timer" that benefits the user but it would be catastrophic for servers (which is where HPET is king)
HPET should have no business in gaming PCs, because the only way to force your entire system to use the more efficient TSC is to disable HPET entirely.
HPET also affects the mouse movement feel. I'm not entirely sure why this happen but if I was to guess I'd say it is because of the "balanced" or "fixed" ratio of interrupts instead of the chaotic nature of TSC, HPET is the Vsync of system timers basically.
HPET's benefits are hardly limited to server clusters and most of HPET's problems are highly dependent on the quality of implementation. Overhead is much higher on HPET than TSC, but it's still going to be irrelevantly low on the majority of modern consumer systems.
The difference between useplatformclock true and false (or with the entry removed) on my more recent systems is so small that it's hard to measure when it comes to DPC/ISR time, or anything else for that matter. Only synthetic benchmarks that specifically poll the timers can reveal any clear difference.
Originally Posted by Chaython
Do I win highest latency?
Page fault resolution time is how long it takes something that wasn't found in memory to be retrieved from elsewhere. It has essentially nothing to do with DPC latency.