Blur Busters Forums

Hello Guys,

i recently installed Windows 10 2009 (Upgrade from 8.1).
I already saw in task manager that it uses more RAM and CPU. What can i do?

There are quite a few things you can do.

Rather than list everything, you can start with some of the basics from Calypto's guide:
https://docs.google.com/document/d/1c2- ... sp=sharing

However, as he mentions, do not enable HPET if you want low latency. I highly recommend to ensure it is disabled for most gamers.

Obviously, read all of the Blurbuster articles. Individual game settings and tweaks can also make a huge difference!

howiec wrote: ↑

27 Oct 2020, 14:08

However, as he mentions, do not enable HPET if you want low latency.

"Better DPC latency" doesn't necessary mean "better input lag".

On my Ryzen PC with Windows 10 v2004 I have HPET enabled in BIOS and "useplatformclock true" in OS because it gives me better mouse input (less input-lag), and honestly I couldn't care less if this tweak raises DPC latency from 30 to 50.

Fr33thy is pretty good:
https://www.youtube.com/watch?v=WxrN-um ... UAHw99aGWW

I did a few things in bios and os

ffs_ wrote: ↑

29 Oct 2020, 12:13

howiec wrote: ↑

27 Oct 2020, 14:08

However, as he mentions, do not enable HPET if you want low latency.

"Better DPC latency" doesn't necessary mean "better input lag".

On my Ryzen PC with Windows 10 v2004 I have HPET enabled in BIOS and "useplatformclock true" in OS because it gives me better mouse input (less input-lag), and honestly I couldn't care less if this tweak raises DPC latency from 30 to 50.

Sorry, but generally speaking, that's incorrect.

Lower max ISR and DPC latency by definition is an indicator of how quickly your system can respond to any given interrupt and execute the driver routine and/or other code.

Clearly, if any of your input devices or GPU processing latency is unaffected in this regard, then technically, you may not see reduced input lag.

However, with testing across 2 systems (my old Q6600 and my current 8700k builds with various GPUs) I've clearly found that HPET has severely degraded gaming performance on my Intel systems.

The "mouse feel" will likely change with HPET on/off so although I don't have any experience with an AMD or Ryzen system, I can guess that you simply prefer the HPET mouse feel in your system-specific setup (HW & SW).

Below are 3yr old FC5 benchmarks with my older configuration. HPET/useplatformclock was literally the only change between the 2 runs.

HPET/useplatformclock OFF:


HPET/useplatformclock ON:

howiec wrote: ↑

29 Oct 2020, 16:16

Hpet generally forces the timer resolution to be rounded down, what it means is that 1 second is not 1 second according to your machine (it's generally slightly less), this screws any form of time sensitive fps or latency calculator, from fps benchmarks to even AIDA64 latency benchmar.

250 frames in 1000ms = 250fps, 250 frames in 950ms = 263fps (just to make an example of how HPET OFF can screw benchmarks in windows 10)

Saying that hpet reduces input lag or increase fps is misleading, I highly recommend to watch Fr33thy video about HPET, the general rule is ON in the bios and OFF in Windows 10.
I find that using this in combination with setting a system 0.5ms timer with ISLC gives a very consistent and low input lag experience.

It must be noted that on some very old computers benefit from hpet OFF in both BIOS and WIndows 10, but these are rare cases, processors nowadays can handle hpet just fine.

axaro1 wrote: ↑

29 Oct 2020, 16:45

Hpet generally forces the timer resolution to be rounded down, what it means is that 1 second is not 1 second according to your machine (it's generally slightly less), this screws any form of time sensitive fps or latency calculator, from fps benchmarks to even AIDA64 latency benchmar.

250 frames in 1000ms = 250fps, 250 frames in 950ms = 263fps (just to make an example of how HPET OFF can screw benchmarks in windows 10)

Saying that hpet reduces input lag or increase fps is misleading, I highly recommend to watch Fr33thy video about HPET, the general rule is ON in the bios and OFF in Windows 10.
I find that using this in combination with setting a system 0.5ms timer with ISLC gives a very consistent and low input lag experience.

Those 3yr old benchmarks were on my 8700k + 1080 Ti.

Not only did I see reduced performance across all games, I also noticed erratic mouse cursor movement and even Windows GUI performance degradation.

So in my case, it was 100% not just some "benchmark timing calculation inaccuracy due to clock timer reference".
Don't confuse clock resolution with the cost of using a certain clock source.

Also, on some systems like mine (Asus Hero X), you cannot disable HPET in the BIOS. Unless I'm missing something, disabling HPET in dev mgr, deleting useplatformclock, enabling useplatformtick should achieve the same/similar results.
FYI, I also enable disabledynamictick .

I watched Fr33thy's HPET video and he's basically saying same thing. Disable HPET/useplatformclock in OS.

Also, to clarify, HPET incurs more overhead compared to TSC. That is the main reason for reduced performance...

Keep in mind that you want to disable all CPU power-saving features in BIOS to also eliminate any effects on latency and the TSC timer.

Keep in mind that the HPET on/off can also distort the calculations used by the benchmark (e.g. show worse benchmarks for something that's benchmarking better to human eyes, and better benchmarks for something that's benchmarking worse to human eyes). Make sure whatever benchmark you use, isn't screwing its own benchmarking up, from the sheer HPET issues itself, mis-benchmarking itself. This does not always happen but worth mentioning that there's (over the last 10 years) been crappy benchmarking software that adds noise to the noise.

Chief Blur Buster wrote: ↑

29 Oct 2020, 18:19

e.g. show worse benchmarks for something that's benchmarking better to human eyes, and better benchmarks for something that's benchmarking worse to human eyes). Make sure whatever benchmark you use, isn't screwing its own benchmarking up, from the sheer HPET issues itself, mis-benchmarking itself. This does not always happen but worth mentioning that there's (over the last 10 years) been crappy benchmarking software that adds noise to the noise.

Agreed. Proper test/measurement methods that also account for various types of error/noise, and careful statistical analysis are often lacking even still today...