Another look at HPET High Precision Event Timer

If you have no problems in games then do not bother.

 

Have tried all energy settings known in windows and the uefi including power saver profile and pci-e link state power management at maximum.

Only ones that caused me performance degradation where there are no known issues is using power saver profile and forcing a certain platform timer (if the OS is already truly choosing the best one).

Timer resolution, C-states/speedstep, core parking, link state power management, gpu performance state (I see people forcing this globally. Prefer Maximum Performance should be a game specific case. Else you waste power for games that do clock the GPU correctly) can all be enabled/left default/set to max and you should be fine (provided your system has no faulty software/hardware).

 

No you don't waste power, because half the time if you DON'T with a GPUBoost card, the game won't keep the clocks stable. It will arbitrarily determine that it doesn't need the performance, and will downclock the card instead... doing what? Causing performance problems!c1:

Let's not even get into GPU Boost, GPU Boost, when boosting has a hard time keeping the voltage stable. At least that has been mine and several others experiences. Prefer Max Performance will at least make sure the base clock is kept as a minimum. Adaptive ****s that up.

With the latest drivers, card clocks correctly adapt when not using Hardware Acceleration(ex: Steam running, even when using Prefer Max Performance. (My 980 is currently idling at 135mhz with Steam open and my web browser open)

 

I chose to mod my bios for that reason. The idea that the gpu knows best just doesn't work all the time, even if a game is maxing out on core boost, there are still power limits going on that can lower your performance without you wanting it.

 

Like intel's speedstep, Nvidia's performance states (GPUboost cards are basically a "many P-state" solution for efficiency. Fermi has a basic 4) only change based on the amount of work the GPU is receiving from the CPU. The GPU driver intelligently knows what performance state it should be at determined by the threshold of work from the CPU. If you set to Maximum Performance at all times you are wasting power for no additional load (some games dont send work efficiently through your GPU and this setting can be for compatibility).

As for instability, what you're figuring out is how the boost table works in GPUboost cards. Some vendors' factory overclocked cards aren't programmed correctly in their video bios. They modify the boost levels but forgot to compensate voltage for those higher boost levels/steps/p-states. I had to edit my GTX 980's bios (maxwell bios tweaker) or else the driver would crash (solid color screen) around 1266mhz due to the voltage (1.013v) not being corrected in the factory (i have it at 1.1v at that boost level). So i had to bump every state after 1266mhz up by about .075mv due to the lack of attention from the vendor (EVGA).

So if you are a crashing on a lower p-state, its more likely an issue with your video bios, not the control panel p-state setting.

 

guys interesting thread. I am in a odd situation tho.

Using the pc clock frequencies tool, I discovered hpet on is better one way, but worse another way.

So queryperformancecounter call is way lower on invariant tsc like most have discovered on new hardware. 5.9ns vs 700ns.

However with hpet on my timegetresolution is halved vs invariant tsc, hpet value is 1000ns vs 2000ns on invariant tsc.

Since I dont know which value holds more importance in terms of performance my situation is not black and white.

games seem smoother with hpet off, and benches score a bit higher. But loading applets in windows is faster with hpet on, paint loads instant with hpet on, but has a small delay with it off.

bizzarre

Ok I got the timegetresolution down to 500ns now using the tool from this post, thanks to the poster. http://forums.guru3d.com/showpost.php?p=5042181&postcount=981

So I am keeping HPET off and using that tool. Win8.1 pro

 

Just keep it on. Windows automatically picks the best available timer. Timer resolution is not affected by the presence of a HPET.

 

Testing it extensively...

Its not worth it if using windows 10. leave hpet on in bios.. leave the OS bcedit alone.

bcdedit /set useplatformclock true ( this cuts performance in half on intel processors and makes performance worse)

bcdedit /deletevalue disabledynamictick ( this one is a troublemaker, it boosts performance slightly but makes the OS slightly unstable)

bcdedit /deletevalue tscsyncpolicy ( Ridculous. Unstable. Leave it off)

 

best guess- leave it alone unless on modern hardware as older tech wasnt fully compatible to request, interrupt or flush at a faster clock/timer.

 

Same thoughts exactly

 

Also if my lines of code is incorrect. My bad but i believe you should understand what i'm trying to say. Refer to this link for more information..https://tweakhound.com/2014/01/30/timer-tweaks-benchmarked/

Its worthless to tinker around with these tweaks. it provides no performance boost, maybe the bcdedit /set useplatformclock true for amd cpus under certain odd scenarios.

But as far as i've tested its worthless with a intel cpu.

 

I came to a similar conclusion, Windows is slightly faster with HPET on in certain situations, like logging in (shaving off 1 second on SSD) or better responsiveness in multitasking when disk thrashing. Didn't notice any difference in gaming.
Funny thing is I didn't pay much attention to this, but with most Windows 10 updates it keeps defaulting back to off and after a while working in Windows I notice a slight lag compared to before only to discover in WinTimerRes that HPET is off again.

 

Omfg, not this again.... -))

 

soo, for best dpc latency, it should be off? ?

(windows 10/z170 pro gaming/i7 6700k/gtx 1060)

 

It doesn't affect dpc all that much. It gives the mouse a much more 'raw' feeling (as opposed to smoothing I suppose), and will probably give you like 1-2% lower performance. I have it off.

 

out of curiosity, why ryzen benefits from disabling hpet?

 

(Not this again!)

Is it confirmed solid info?

 

http://imgur.com/a/gxcnG

 

Maybe he misspoke in other points as well?

 

i trusted germany article by computerbase.de rather that those stupid youtube reviewer like jayztwocents, paul h/w, kyle bitwit, linustechtips etc.

"In addition, Hallock pointed out that testers might have wrongly set up their motherboards or Windows - both were not the case at ComputerBase. Also, the High Precision Event Timer (HPET) was disabled under Windows."

credit to Volker Rißka @ computerbase.de
https://www.computerbase.de/2017-03/amd-ryzen-1800x-1700x-1700-test/