but how far do you want to go?
Hey, nice what you're doing here. It's really solid.
You should set up a Discord server for your project, easier to get feedback and talk about general tweaks, optimizations, commands etc.
Also, wanted to mention something:
This breaks certain Store applications, meaning they won't install.
Examples: Asphalt 9 and Forza Horizon 4.
I guess they need the encryption somehow?
I suggest you remove it, in case someone in the future wants to download bigger apps through Store, and personally, I haven't seen any performance impact with it NTFS encryption enabled. It's not like it's BitLocker.
hi mt not sure if you checked this guide or no i cant confirm it tho since he didnt talked about exact values wanna copy paste it here :Will do, and yes I've seen that video!
Mostly interested in his document, I must say it was merely to test but HPET is slow and on later Intel boards even detrimental (Like on my Z270G). So I tested his invariant TSC settings and it resulted it way better mouse polling and back to old W7 0.500/1.000 timer. From my former testing way back, W7 was behaving best with in-engine fps limiters which also showed 0.500/1.000ms. (FPS limiters were much less fluctuating).
Everything feels incredibly snappy. I've turned off my ISA bus completely pretty sure HPET falls under that too though.
yeah i saw this before https://www.overclockers.at/number-crunching/the-hpet-bug-what-it-is-and-what-it-isnt_251222/page_2Yeah I know this guys github, has a lot of useful things. Some might be hard to confirm, other things I don't entirely agree with either.
My stuff in profile is defo not hard evidence, and I'm still in the process of testing a lot of it more in-depth.
But one thing is clear, timers affect in-game fps limiters. And you'll mostly want to use these for reasons way beyond the scope of this thread! Forcing HPET on the other end can have detrimental effects on system wide performance.
The fact that 180x+ builds now all use 10Mhz synethetic QPC and claiming nothing have to/can be changed, doesn't mean the clocksource implementation behind it is no longer relevant. This is probably a flawed assumption as if i would to force HPET right now, behind this synthetic timer, with the HPET Intel platform quirk I can bet things would quickly go down hill
Interesting read perhaps.
thanks buddyWell I never felt a noticable decrease performance vs 1607.
As for the 0.500/1.000, the values happen when using useplatformtick yes, seems to be more stable compared to the non-rounded numbers 0.496 etc which are off. More like Windows 7 as thats the only OS i get rounded numbers and had the least jitter.
Whether its 0.500 or 1.000 or even 2.000 does't matter as long as they show proper rounded values (ending with 0)
I can't make recommendation for IRQ assignment, but maybe set priority for yourself. I think mouse polling is really important, but core 1 is always highest jitter due to kernel running on core 0, so I put usb on core 1 without sharing.
That is just my idea, let me know if you have better theories
Especially useplatformclock/tick. Could be highly platform dependent.Hi!
Bcdedit entries can work for you or not. Might have to experiment ;-)
Especially useplatformclock/tick. Could be highly platform dependent.
I mean without sharing that you serve all usb controller interrupts on cpu1 and all other hardware on cpu 0,2,3 (The more cores the more options possible)
cpu0 probably has most jitter due to kernel/drivers running on it, so also adding interrupt serving to it can really impair polling. A core can only do one thing at a time synchroneously so the more jobs the more jitter/interference.
Redhat Linux also has some nice docs on the web about reducing latency/jitter altho it is based on linux, the basics are similar.