Another look at HPET High Precision Event Timer

What happens if you disable HPET in DM and reboot with useplatformclock=yes?

 

hi, im using gaming laptop and what should i do with it

 

Yeah, this idea visited me too after I replied to Astyanax .Some day I will test it. May be...

Enjoy your games. No need to do something if you are satisfied with performance.

Update: Anyway, you will not find viable tweaks in this thread.

 

Read the manual. Until then, make sure never to stare at it for more than 2356 consecutive seconds (serious, no!) but more importantly: under no circumstance (doesn't matter how hard it pleads) shall you allow it to breath in any hexamethanolmonohydrate salt through it's cooling fan (no, you don't want to know what happens!). Ah, and never let it alone with kids, elderly or anybody else who might be susceptible to addiction or manipulation.

 

i saw many bcdedit configs here, so anything stable with my laptop xD
i'm using 1709 btw, sometimes 1909

 

All these configs are for legacy hardware. On modern hardware you should not touch BCD settings.

 

@mbk1969. Did you test the Windows Keyboard Languages/ mouse lag input bug?

US Keyboard caused me lag input, fixing it when changing to Candian or UK . but I need to go back to US because keyboard buttons don't work as intended as US on the others.
Though testing now with UK it works fine. so I don't know if it's a bug that just changing it at first fix it and reverting back still have it fixed.

I saw someone on Overclock Thread or somewhere else saying that.

Did anyone with Lag Input try to do that? Maybe we just messing with timings and stuff and eventually, it's something in the OS. Because it was so random though I always had weird mouse smoothing and it was fixed exactly at the time I changed Keyboard Language, removing US. I even had a weird Mouse Lag as soon I removed the US keyboard. it was the weirdest thing I ever saw, it's shouldn't even be related, but at the moment I had fun playing games after the weirdest time I had this lag input issue. for almost 4 years I couldn't play games with this lag.

Also, the whole story is that this Windows I am using now was used on Intel puter on a Harddisk. Then I cloned it to SSD using the Samsung migration on 850 256Gb. Then I cloned it to Sandisk Extreme 480 XPS GB SSD. Then I bought AMD 1950x puter and I used the Sysprep to move it. everything worked fine as it is. But this bug was since windows 7. and it still happened to me on a the old Intel puter with HDD on 125 Microsoft mouse on new windows 10 install.

 

it'll fall back to APIC timers.

 

Okay so a wierd thing. When i have hpet enabled in devicem Snappydriver is fine using the orginal driver for hpet, which is none according to the devicem. But when i disabled it, Snappy finds a never version?

Can anyone explain why this happens? And should i have the new one?

http://www.bilddump.se/bilder/20200117095523-79.136.70.78.png

 

I suspect you have overtweaked your Windows.

Just newer inf-file with newer version string.

 

It seems there is big issue with 1903 and 1909. DPC issues is not only for me. it's random but it happens.
Something is just weird, it's not the OS. could be Nvidia drivers or just something in the windows. I never had that issue because of my computer. it always was Windows issue.

EDIT : I think the issues started since I Enabled back HPET in the bios. With it disable now the mouse is much better. I also Enabled IOMMU and SR-IOV which I disabled before. I also changed the memory to procODT 60 instead 48.

It's not CPU,GPU,RAM, hardware issue. it must be some setting that needs to be set specific.


Because when the Freeze happens in the COD MW 2019 it only freeze in the game. sound that I played using Potplayer kept going. so this is the weird thing. it's something related to the game itself.

I don't have any BSOD or errors.

I will continue to investigate.






It's randomly goes to Storport.sys or wdf1000.sys or ntoskrnl.exe. It never ever was like this. and I did not touch any weird tweaks or removed things. I just disabled telemetry and others.
Also a lot people said the same for 1903 and 1909. 1809 worked fine.

Else it could be my eClaro card is causing the issue. drivers are old.

 

try the 422 hotfix driver

 

I tried, still having minor freezing and stuttering when Latencymon is not open, but when it's open I have freezing of 5-7 sec. it shouldn't freeze like that even when latency mon is opened. I mean I have 16 cores, the game uses only 8+8 SMT. so I have free 8+8 SMT cores available and the game doesn't do loading in mid-game, it cache everything every time installing new GPU drivers.

 

I've found why there is high latency with in-depth. it's the GPU power saving, it affects the numbers. I think it's related to how windows render it. so one secret is out of the window.. it's weird that when HT/SMT is enabled it effects it like this. with HT/SMT disabled it doesn't happen.

Windows XP wasn't like that so it didn't happen nore will be. because it doesn't have pretty windows to render, doesn't go to a power-saving like now and should speed up when anything happens on screen.

Now even weirder thing. NVidia has set Adaptive to Microsoft Shell Experience as default even if you set anything else on Global, it would stay Adaptive. They did it to some other files also.

Also the issue with the CPU affinity I fixed it by setting it only to Core0 and now all on Core0. but it's weird, there must be something in the registry that didn't get cleaned up. and I don't know what to remove.

 

its absolutely normal for the gpu to increase dpc when changing power states - the card doesn't respond to requests for the cycle it takes to switch from high to low or low to high though i don't know the asic enough to explain whats happening in this cycle to you.

it definitely shouldn't spike to over 1k during this transition though, almost always when this happens another device is an offender, reminder that all dpc's are queued and while they can be distributed to more than core 0 to process, higher priority requests will go to core 0 which is just about never in C3 or higher.

It seems a little off for nvidia to set that process to adaptive, since it will mean Optimal is never used.

 

I never really understood the difference between Adaptive and Optimal. The text in NCVP suggests something like there are only 2 states: 'Adaptive', 'High Performance' and Optimal selects one of these automatically. Guessing purely from their names: Adaptive is dynamic clocking but with some kind of detection to go into fixed max P state (by somehow recognizing demanding games and such - by process names or "adaptively" from the utilization profile), Optimal is strictly dynamic (clocks are always throttled on the fly whenever it looks possible with no significant performance hit), High Performance is always evidently fixed high clock (well, base clock, not necessarily "turbo boost" clocks).
I am also not sure about how the driver selects things like this when multiple processes are running with different selections. The obvious guess is window focus but even that can be tricky at times.
But shouldn't these clock changes happen in microseconds rather than milliseconds anyways?

 

Optimal puts the GPU into a sleep state and renders from the last frame in the buffer when nothing new has been rendered, it is otherwise the same as Adaptive when actively rendering.
Adaptive continues to render new frames even when no updates are occuring.

If it were as you say, then the moment you open a web browser of any vendor you would not see some games getting drops to low power 3d as all the browsers are configured to Adaptive in the profiles.

No, both adaptive and optimal go down to low power 3d and lower based on gpu util %, its usually promoted to 3D at 50%, and drops to low power 3d once util goes lower than 20%, and continues down to idle if below that.

The driver is also tuned to know that the clock speed directly affects how UTIL % is calculated.

Yes, but in that microsecond of nothing being processed the queue backs up like a single lane left turn that a semi has just come by in the lane it opens out to.

 

The point is 139Mhz is not enough for Windows graphics/Window and power saving was not in XP, and we already know power saving causing serious issues. it could be best for office users and web browsing, but not for gaming.

This latency you are seeing in the In-depth tests is about how your latency to lag input. you wouldn't want it ever to go above 30 on normal usage, and it doesn't changing power state because it stays 139mhz. and for normal dekstop usage it is annoying, it's not windows XP without graphics.. that's why it worked fine. some things required GPU so why do I have to have latency while using desktop?? with /AMD GPU it doesn't happen.

 

just to add

from my readings/experience on linux, basically tsc is almost always preferred now, since accessing HPET is an io operation, it is impacted by the spectre/meltdown mitigations , which causes a performance penalty when using it. avoiding the use hpet entirely should be the default for both linux and windows now, unless like you said the tsc timer is buggy(this can be worked around on core 2 platforms by disabling c states higher than 2),

older platforms have HPET on the Southbridge(ich5-10? and nforce chipsets on intel i think) on modern chips its all integrated on the cpu (apparently on the pch now), so access times shouldn't be that bad.

there is also the acpi power management timer ( 3.58 MHz) which pretty much all platforms that support acpi have, I dont know for sure if windows uses it, but it might since ive seen some clock tools report this 3.58 mhz as the tsc timer frequency, on a motherboard i have that lacks hpet support and has broken tsc.

 

HPET is still on the PCH with modern systems.