revisiting hpet bcdedit tweaks: what are your timer bench results and settings?

[zoulztealer]

only for advanced users: be aware that once settings are defined by bcdedit these can cause irreversible changes and damage to your system such as low fps and stuttering until windows is reinstalled!

currently im in the process of revisiting some supposed tweaks for hpet and their implications.

the theory is that hpet can cause some issues like stuttering, high input latency or low fps performance.

some systems, games or apps may also run better with hpet enabled, some may run better with hpet disabled.

the tweaks in question are:

useplatformclock
useplatformtick
disabledynamictick
(tscsyncpolicy)

some guides consider disabling hpet by the following commands, while other guides even consider this as re-enabling hpet.

bcdedit /deletevalue useplatformclock
bcdedit /deletevalue useplatformtick
bcdedit /deletevalue disabledynamictick
bcdedit /deletevalue tscsyncpolicy

i argue that this is not to be set equal with disabling hpet since it only removes a definition that may or may not have been previously defined.

if the bcedit value gets deleted i assume that the setting falls back to what is defined as per bios setting.

i argue that to disable hpet following commands would have to be used and most importantly hpet disabled in bios (some guides even argue that these settings only have effect if hpet is kept enabled in bios):

bcdedit /set useplatformclock false (this setting should also have a specific effect when set to true and hpet is disabled in bios)
bcdedit /set useplatformtick false - or - bcdedit /set useplatformtick true (microsoft documentation says that enabling this setting leads to disallowing of synthetic timers, so i believe to properly disable hpet useplatformtick would have to be set to true if hpet is considered a synthetic timer)
bcdedit /set disabledynamictick true

thread connects on a lot of issues:

useplatformtick and/or disabledynamictick on mouse input lag - Page 3 - Blur Busters Forums

be aware that once settings are defined by bcdedit these can cause irreversible changes and damage to your system such as low fps and stuttering until windows is reinstalled!

so it is quite unclear what the respective settings each do and how they must be arranged in combination in order to achieve full enable- or disablement of hpet.

then yesterday i stumbled upon timerbench which is a great tool that not only displays all kinds of useful information what timer is used but also provides a platform for benchmarking the system and detect if timer related issues are present:

The HPET bug: What it is and what it isn't

Anandtech recently released an article that pointed out problems with their CPU reviews due to an enabled High Precision Event Timer in Windows. Some Intel processors suffered from decreased performance in games and other benchmarks. Since then a lot of misconceptions are going around. People...

www.overclockers.at

i have yet to contact the author of timer bench to find out what methods and commands he actually exactly used for disabling hpet and how the status is determined.

my system seems to run perfect with hpet enabled in bios and forced on with bcdedit settings:

listing of my bcdedit entries:

systems that are affected by the hpet bug would result in low gpu load / utilization or other issues that would likely become evident in the benchmark result, as was illustrated in the article from overclockers.at about timer bench; for some reason the cpu is not able to fully utilize the gpu resulting in low fps (hpet gpu load ~44%, itsc gpu load: ~96%):

so my theory is that timerbench can be used to detect whether a system is affected by the hpet bug by showing low gpu load or other issues - and - when a system is non-affected there likely wouldnt be any issues visibile in the results of the bench.

now i am curios as to what are your bcdedit settings and what are your timerbench results and statistics?

if you are capable to do so, please post a screen of your bdcedit settings and your results in timer bench as well as whatever you know about the hpet issue.

be aware that switching hpet on / off with either timerbench or manually by bcdedit can break your system and make it unbootable!

 

[Solaris17]

now i am curios as to what are your bcdedit settings and what are your timerbench results and statistics?

Wait till you take into consideration not all HPET hw implementations are equal.

 

[zoulztealer]

Wait till you take into consideration not all HPET hw implementations are equal.

yes, i have already suspected this. hpet should be different on every mainboard / cpu or something like that. and also different apps and games use different, conflicting timers.

im using useplatformclock and useplatformtick enabled at the same time and i think this is not alright because those two settings are conflicting but im not sure.

 

[chrcoluk]

If I remember right isnt the reason we have faster more modern timers due to HPET having high overhead? as I understand it the default is for HPET to be available for apps that want it such as hwinfo, but not the default timer.

 

[ShrimpBrime]

I use the "useplatform clock" yes for the following hardware bot rules since release of Windows 8

Quoted
"Due to severe validity problems with the Windows 8/8.1/10 real time clock (“RTC”), not all benchmarks results achieved with Windows 8/8.1/10 can be trusted. The main problem lies with the RTC being affected when over- or under clocking under the operating system. The operating system uses the RTC as reference clock, and benchmarks use it to reference (benchmark) time.. The BenchMate benchmarking tool enables users with affected setups to safely use Windows 8/8.1/10 operating systems. This tool verifies the RTC and on top enforces enhanced security for all the benchmarks included in the BenchMate suite."

Figure this was an easy way to explain Why to enable HPET. Doesn't matter what hardware, it matters what operating system. I am on X58 right now for kicks, using w10, it's enabled.

 

[zoulztealer]

useplatformtick and/or disabledynamictick on mouse input lag - Page 5 - Blur Busters Forums

in this thread they are also talking about the timers being in sync or not when tested with gpuz built-in timers tool. my timers seem to be pretty much in sync but i just made a screenshot and the rtc clock seems to lag behind sth like 0.1 sec. let it ran for 2-3hrs now and also 0.1 again.

Surprising Observation! HPET On vs. Off - Page 3 - Blur Busters Forums

Melody's Ultra Tweaks Pack - Misconceptions about timers (HPET, TSC, PMT...)

What do useplatformclock and useplatformtick do? useplatformclock disables TSC and uses the platform source clock instead (HPET or PMT). PMT is used when HPET is disabled in BIOS. useplatformtick disables TSC tick and uses the platform source tick instead (RTC). Does disabledynamictick work when

sites.google.com

specifically these kind combos would have to be tested and diagnosed with timerbench. i think the effects and results will be different with every system.

 

[chrcoluk]

Useful explanation, some boards dont allow disabling HPET in the bios (mine included), so PMT might be locked out for some people now, and as expected it does say forcing HPET on for everything is a bad idea.

 

[DaemonForce]

I wasn't aware of any of these settings and the whole system is probably the same.

What setting am I using?

 

[zoulztealer]

I wasn't aware of any of these settings and the whole system is probably the same.
[...]
What setting am I using?

seems like you are perfectly running itsc and hpet is disabled. everything is exactly as it should be.

Useful explanation, some boards dont allow disabling HPET in the bios (mine included), so PMT might be locked out for some people now, and as expected it does say forcing HPET on for everything is a bad idea.

yeah, but im still not entirely convinced of melodys and others analysis as well, because they seems to be too much generalized and dont take into account the settings and results can be different on every machine. it seems though that disabling hpet delivers consistently higher fps and better gpu utilization throughout all benchmarks. also melody calls placebo effects on lower latency which isnt true since this definitely does have a noticeable improvement on input latency and smoothness.

i find more reason in what is said in this thread, the results posted there and what the developer of timerbench has to say:

TimerBench: Ein Benchmark für Windows Timer

Präzise, verlässliche und performante Timer sind die Grundlage für einen reibungslosen Betrieb von Windows, insbesondere PC-Spielen, Benchmarks und Multimedia-Anwendungen. Trotzdem wird das Thema selbst von Enthusiasten kaum gewürdigt. Jeder kann heutzutage seine PC-Komponenten übertakten, um...

www.overclockers.at

english version:

The HPET bug: What it is and what it isn't

Anandtech recently released an article that pointed out problems with their CPU reviews due to an enabled High Precision Event Timer in Windows. Some Intel processors suffered from decreased performance in games and other benchmarks. Since then a lot of misconceptions are going around. People...

www.overclockers.at

im still waiting on an answer from the dev what commands he uses to consider hpet dis- or enabled but the only proper way so far to me seems to be using the timerbench app for setting and diagnosing of the hpet status.

i dont understand how according to melodys tweaks this will "force on" tsc, because if only the bcdedit values are deleted and hpet is enabled in bios, the os will probably use hpet:

TSC + TSC without desync: bcdedit /deletevalue useplatformclock - bcdedit /deletevalue useplatformtick - make sure HPET is enabled in BIOS

im also using custom timer resolution from islc of 0.5 and i can definitely notice every step of granularity. the lower the timer the smoother the machine becomes. so i dont really understand the recommendation to not use external timer apps anymore since past win 1905 neither.

A Timely Discovery: Examining Our AMD 2nd Gen Ryzen Results

www.anandtech.com

The interesting thing to come out of both Intel and AMD is that they seem to not worry if HPET is enabled or not. Regardless of the advice in the past, both companies seem to be satisfied when HPET is enabled in the BIOS and irreverent when HPET is forced in the OS. For AMD, the result change was slight but notable. For Intel we saw substantial drops in performance when HPET was the forced timer, and removing that lock improved performance.

Mesa Fixes A Large Performance Regression For Systems Using HPET - Phoronix

www.phoronix.com

The High Precision Event Timer (HPET) has long been a source of issues for Linux developers and it turns out systems relying on HPET rather than the CPU's TSC have in recent months suffered significant performance degradation with the Mesa OpenGL driver code.

Opened back in April was this Mesa bug report around a "huge performance regression". and was tracked to being due to a Mesa OpenGL threading "glthread" change. The performance could be as much as halved since the regression was introduced.

The problem was ultimately found to be when using Mesa OpenGL threading while making use of the High Precision Event Timer (HPET) rather than the CPU's Timestamp Counter (TSC) for hardware timing.

 

[chrcoluk]

For me HPET just has too much overhead so the argument would be PMT vs TSC, and similar for the tick old vs new.

Older games were developed around PMT so may work better when using that clock source, but of course new software takes preference, and as I understand it new games are designed around TSC, and on top of that if you think about hardware that has very high levels of polling I believe TSC is optimised for that as well.

For tweaking an existing system, I think disabling dynamicticks is pretty much harmless to do, might be some gains by reducing wake up latency, everything else though I think is a case of it might work out ok, or it might make things worse, I would only consider force enabling the platform clock if the bios was configured to PMT. I would not enable useplatformclock if bios is set to HPET. The ticks from what I understand have a smaller effect than the clock configuration so even if you get a regression with RTC ticks, it shouldnt be anyhere near as nasty as HPET.

Some people disable useplatformclock which is actually different to the undefined behaviour, when it isnt defined, it will allow applications to request and use HPET specifically, such as hwinfo and some media apps. Some people dont like that is possible as potentially a background app requesting HPET could stutter a game so they disable useplatformclock, what I am not sure of though is if you disable useplatformclock will it prevent TSC sync?

I wasn't aware of any of these settings and the whole system is probably the same.
View attachment 361916

What setting am I using?

Undefined, which is the default, assuming bios is also default, it means you using TSC clock with TSC tick, and have TSC sync enabled. Apps will be able to use HPET if they request it.

 

[zoulztealer]

there should be benchmarking specifically with the dynamic tick setting and tsc sync policy being switched. theoretically there could be a lot of different combinations of settings and effects / faculty.

maybe it would also be important to delete all respective bcdedit entries before running timerbench since its unclear if and how the settings in bcdedit and bios are affecting the results. proper use would be enable hpet in bios and fresh windows install without any bcdedit mods before running timer bench first time. or install windows fresh with hpet disabled in bios if you want it hard disabled.

also if you have bcdedit commands already present probably best to delete them, reboot, etc before making changes in timerbench. but there are always settings already defined but which are not visible to the user in the default bcdedit enum (i read in one of the blurbuster threads)!

highest fps solely alone is not the most imporant factor, but to me that is input latency and absolute smoothness and snappyness, no stuttering after everything has been loaded and cached. fps can be high but jittery; optimization should be done in the direction where you lose a little bit of max fps but gain better fps stability, smoothness and input latency aka "ryzen effect" or ssd, ram / cache effect or effect of having much or more cores and hyperthreading, threadripper, effect of well optimized game / consoles, etc.

i wonder what happens if you install a modded or even updated driver to the hpet device in device manager. the default / ms driver is from 2006, kinda old. maybe there are improvements to be made.

 

[Apocalypsee]

I'll try on my laptop when I got back home. Though laptop usually it's something else like WiFi cards. My laptop exhibit like milisecond stutter with drop in audio every few random minutes. Still, worth checking.

 

[ShrimpBrime]

HPET must be enabled just to run X265 Benchmark. You can (link is from HWBot obtained from the benchmarks DL list) DL it from here.
If HPET is not enabled, this benchmark will notify as such and tell you how to enable it.

If your system has any cpu or memory instability, it won't complete. Similar demands to Y-Cruncher, but you can run over-kill mode and a couple other tweaks to improve CPU FPS.

I DO wonder if some games actually do run better with HPET enabled or disabled.

Almost all motherboards are enabled by default in BIOS, however the Operating system is NOT enabled by default. I suppose a research to why Microsoft has this disabled. I'm too lazy, but the system doesn't seem to respond any differently and all apps and programs seem to run the same, similar or close enough where just can't tell the difference, but I'm certain some programs and apps would benefit either enabled or disabled.

Don't have spare time to check, but I just done a fresh W10 install without updates and HPET is in fact disabled by default, I made sure to confirm it. All you have to do is try and open/run X265 benchmark... Using new motherboard Asus TUF Z790 Gaming Plus Wifi and slightly ab-used 14700K onto Samsung Pro 980. All system defaults including memory 2400mhz 40-40-40. CBR20 13636 max temp 77c at 266w peak Max VID reported 1.359v Using Bios 1604 ME version 16.1.30.2307v4. NOTE: I don't know which ME version has VID spiking 1.50v++, but I have not witnessed it on this older bios and ME version, but this is the first day, I haven't any conclusions yet. So far, the system doesn't "Need" HPET enabled that I can tell unless I'd like to run X265 benchmark.

______
TLDR:
HPET disabled by default "fresh" W10 Pro 64 bit install (All default system settings)

 

[Apocalypsee]

I just tested the timer inside CPU-Z on my laptop, it runs very odd, stutter every few second. This is what I got after running just under a minute. I think HPET is disabled by default (its running W10)? EDIT: Just tested it and yes x265 gives warning HPET is not enabled

I run Timerbench but it only detect intel iGPU not the nvidia card. It looks something like this, looks ok I guess?

FURTHER EDIT: Enabling HPET makes it a stuttery mess even in Windows. So I back to disabling it. The milisecond stutter might related to something else, perhaps low power state? Idk but I need further reading

 

[zoulztealer]

I just tested the timer inside CPU-Z on my laptop, it runs very odd, stutter every few second. This is what I got after running just under a minute. I think HPET is disabled by default (its running W10)? EDIT: Just tested it and yes x265 gives warning HPET is not enabled

I run Timerbench but it only detect intel iGPU not the nvidia card. It looks something like this, looks ok I guess?

FURTHER EDIT: Enabling HPET makes it a stuttery mess even in Windows. So I back to disabling it. The milisecond stutter might related to something else, perhaps low power state? Idk but I need further reading

yes, it could be due to power saving features enabled per bios and per windows power plan as well whats causing crackling sound in audio etc. please make another bench with your nvidia gpu. your current bench doesnt show gpu load for igpu. i think timer bench cant read gpu load from igpus.

check in your bios whether hpet is enabled and check in device manager if you have a high precision event timer device under system devices and whether you have any faulty devices with yellow exclamation mark or ghost devices in device manager. could also be your issues will disappear if you uninstall or update your chipset drivers or make a fresh win install with optimized bios settings or there are other 3. party softwares running in the background which are causing this.

when debuggig you want to start with least invasive / most easy to apply methods first and then work you way through to the harder ones.

https://www.uwe-sieber.de/files/DeviceCleanup.zip

make a screenshot of your bcdedit entries too and post it here before making any changes. if your several timers are stuttering this means they are having issues with syncing each other and theres a good chance this can be solved with bcdedit commands or similar.

run bcdedit /enum /v from elevated command prompt or windows power shell admin.

it could also be you will get an improvement already by solely adding disabledynamicticks in bcdedit:

bcdedit /set disabledynamictick true

try another high performance power plan from bitsum process lasso for example and disable every power saving feature in bios and the windows power plan. try running your benchmarks while having process lasso on in the background.

https://bitsum.com/ (process lasso)

ParkControl – Tweak CPU Core Parking and More

ParkControl is free software to display and tweak CPU core parking settings in real-time. It also has an assortment of complimentary power automation features.

bitsum.com

setting a low custom timer of 0.5 with islc is also an option for you:

Intelligent standby list cleaner (ISLC) v1.0.3.4 Released - Wagnardsoft Forum

disable core parking with park control:

dpc lat and especially latency mon can help you find issues with dpc latency and driver:

DPC Latency Checker

Testet die verfügbaren Ressourcen des Rechners zum fehlerfreien Verarbeiten von Echtzeit-Datenströmen.

www.heise.de

for proper diagnosis use latency mon. let it run after your computer booted and loaded everything up and is idling:

Resplendence Software - LatencyMon: suitability checker for real-time audio and other tasks

LatencyMon: suitability checker for real-time audio and other tasks

www.resplendence.com

I just tested the timer inside CPU-Z on my laptop, it runs very odd, stutter every few second. This is what I got after running just under a minute. I think HPET is disabled by default (its running W10)? EDIT: Just tested it and yes x265 gives warning HPET is not enabled

I run Timerbench but it only detect intel iGPU not the nvidia card. It looks something like this, looks ok I guess?

FURTHER EDIT: Enabling HPET makes it a stuttery mess even in Windows. So I back to disabling it. The milisecond stutter might related to something else, perhaps low power state? Idk but I need further reading

essentially try this here (diagnose and setting for latency mon):

Surprising Observation! HPET On vs. Off - Page 2 - Blur Busters Forums

 

[Apocalypsee]

yes, it could be due to power saving features enabled per bios and per windows power plan as well whats causing crackling sound in audio etc. please make another bench with your nvidia gpu. your current bench doesnt show gpu load for igpu. i think timer bench cant read gpu load from igpus.

check in your bios whether hpet is enabled and check in device manager if you have a high precision event timer device under system devices and whether you have any faulty devices with yellow exclamation mark or ghost devices in device manager. could also be your issues will disappear if you uninstall or update your chipset drivers or make a fresh win install with optimized bios settings or there are other 3. party softwares running in the background which are causing this.

when debuggig you want to start with least invasive / most easy to apply methods first and then work you way through to the harder ones.

https://www.uwe-sieber.de/files/DeviceCleanup.zip

make a screenshot of your bcdedit entries too and post it here before making any changes. if your several timers are stuttering this means they are having issues with syncing each other and theres a good chance this can be solved with bcdedit commands or similar.

run bcdedit /enum /v from elevated command prompt or windows power shell admin.

it could also be you will get an improvement already by solely adding disabledynamicticks in bcdedit:

bcdedit /set disabledynamictick true

try another high performance power plan from bitsum process lasso for example and disable every power saving feature in bios and the windows power plan. try running your benchmarks while having process lasso on in the background.

https://bitsum.com/ (process lasso)

ParkControl – Tweak CPU Core Parking and More

ParkControl is free software to display and tweak CPU core parking settings in real-time. It also has an assortment of complimentary power automation features.

bitsum.com

setting a low custom timer of 0.5 with islc is also an option for you:

Intelligent standby list cleaner (ISLC) v1.0.3.4 Released - Wagnardsoft Forum

disable core parking with park control:


dpc lat and especially latency mon can help you find issues with dpc latency and driver:

DPC Latency Checker

Testet die verfügbaren Ressourcen des Rechners zum fehlerfreien Verarbeiten von Echtzeit-Datenströmen.

www.heise.de

for proper diagnosis use latency mon. let it run after your computer booted and loaded everything up and is idling:

Resplendence Software - LatencyMon: suitability checker for real-time audio and other tasks

LatencyMon: suitability checker for real-time audio and other tasks

www.resplendence.com

essentially try this here (diagnose and setting for latency mon):

Surprising Observation! HPET On vs. Off - Page 2 - Blur Busters Forums

I actually did screenshot bcdedit and latencymon when I did that yesterday, I thought I put up but I didn't. I'm busy Googling for answer. The highest latencymon detect is nvlddmkm.sys. I'm going to disable power saving stuff and disable dynamictick (I think I did this, but I can't remember which one I test last night). I also buying Intel AX210 to replace the Mediatek MT7921 Wifi card that ASUS love to use on all its laptop.

That device cleanup tool might be useful for my desktop, as the windows is from windows 7 i5 4670k era that I update periodically, thanks!

 

[zoulztealer]

I DO wonder if some games actually do run better with HPET enabled or disabled.

found another two good articles on that:

Windows Timer Resolutions - Why Disabling The High Precision Event Timer Makes Sense - XBitLabs

The HPET (High Precision Event Timer) is a timer that is used in most PCs and in some cases, it increases performance since it is a very precise timer. However, what was noted by many gamers trying to optimize their systems is that HPET can also cause FPS loss, freezes in games, and higher...

www.xbitlabs.com

Further Clarification on Timer Resolutions and HPET Settings - XBitLabs

As explained in this previous article HPET or the High Precision Event Timer is a suboptimal setting if you are trying to get the best FPS stability and the lowest latency/most stable polling rate. While disabling it and just using the CPU TSC timer is a great start for most there are a couple...

www.xbitlabs.com

As explained in this clarification post you do not need to disable HPET in the BIOS or Device Manager because those changes either do nothing or cause issues by enabling outdated system timers.

Once off, your system will use a TSC clock and you will need to then select between running RTC or TSC ticks, TSC sync policy, etc.

Most custom Windows 10/11 builds already have HPET off so if you get an error message after putting in the command it simply means that HPET was already disabled.

Choosing Between TSC & RTC Ticks
Your system clock has already been put in place (TSC) which means the next important component is to figure out the system tick. You can find more information about the system ticks in the clarification post but the short answer is that you will need to test both ticks (TSC+RTC vs TSC+TSC) to see which one runs better on your system.

RTC is an outdated tick that is not dynamic and has no optimizations which might decrease your system performance and conflict with high polling rate mice. RTC is recommended for unoptimized systems or people who just want the process to be as short as possible without further testing.

Most tuned systems should run TSC+TSC since it is a higher precision timer resolution compared to RTC which was intended for legacy applications (still lower precision than HPET clock).

What Does ”Useplatformtick” Do?
Using the platform tick command will force your system to use RTC (real-time clock) which is an older tick. This is the usual combo that people go for when trying to stabilize their systems and polling rate for 4K-8K Hz mice (bcdedit /deletevalue useplatformclock & bcdedit /set useplatformtick Yes).

This results in your system using the TSC clock plus the RTC timer which is optimal for un-tweaked systems.

What Are Some Recommended Combinations of System Clocks and Ticks?
If you have a modern system with a tweaked BIOS and tweaked/custom Windows install it is recommended you try running TSC clock + TSC tick without desync: bcdedit /deletevalue useplatformclock – bcdedit /deletevalue useplatformtick – make sure HPET is enabled in BIOS.

For those who have a more basic system setup using the TSC clock + RTC tick is recommended: bcdedit /deletevalue useplatformclock – bcdedit /set useplatformtick Yes. This should result in better inputs in games and a more stable polling rate for your mice.

Additional BCD Commands
Besides testing and figuring out which system clock and tick runs the best on your system you should also disable dynamic tick since it is a power saving feature designed for laptops. You can do so by using the bcdedit /set disabledynamictick yes command. If you want to use RTC for system tick you can delete the value altogether (bcdedit /deletevalue disabledynamictick) since RTC is not a dynamic tick counter.

The last thing left to do is for you to test which TSC sync policy will work better on your system: bcdedit /set tscsyncpolicy Enhanced or Legacy. Test with both Enhanced and with Legacy and do different polling rate tests and FPS tests to see if you get better values with one of them, or if there are any changes at all..

I actually did screenshot bcdedit and latencymon when I did that yesterday, I thought I put up but I didn't. I'm busy Googling for answer. The highest latencymon detect is nvlddmkm.sys. I'm going to disable power saving stuff and disable dynamictick (I think I did this, but I can't remember which one I test last night). I also buying Intel AX210 to replace the Mediatek MT7921 Wifi card that ASUS love to use on all its laptop.

That device cleanup tool might be useful for my desktop, as the windows is from windows 7 i5 4670k era that I update periodically, thanks!

some drivers or other files can cause a high dpc latency for a short moment if theyre being loaded first time or busy doing something.

disconnect from the internet, close all background apps and launch latency mon for a short time 5-10s. the most important value is in the first line: current measured kernel timer latency.

mine is around 8-22s, which is already very low. id say value of 100-500 can be quite normal and still ok. but you will see this value decrease if your tweaks have been succesful or not.

highest reported can be quite high and are not an issue if the live measurements are in the green.

example for bad driver causing high latency:

example for good or optimized conditions:

dpc lat should be constantly in the green too and go as low as possible when idling:

since i did the bcdedit tweaks mine always hovers on 500us, sometimes it goes down but only very rare. one time it was very low direct after booting up but then something loaded in windows and it jumped to 500us again and kept hovering there. im not sure if this is due to timer resolution from islc, power plan process lasso or something else but i believe it could also be due to forcing hpet always on and using rtc which is non-dynamic. it must be some kind of setting or app. usually the different points of measurement should always be different and optimally strifing towards to lowest possible value. maybe the cpu is doing constant timer sync / conversion.

my dpc lat chart shows something is keeping cpu busy on a certain level on purpose, its not optimal. it also doesnt rise when playing or other activity much. theres also one decrease in latency while the game is still running (marked with ?). usually the system would go down to this value but something keeps latency high / cpu busy on purpose and latency hovers all time on the same value:

I actually did screenshot bcdedit and latencymon when I did that yesterday, I thought I put up but I didn't. I'm busy Googling for answer. The highest latencymon detect is nvlddmkm.sys. I'm going to disable power saving stuff and disable dynamictick (I think I did this, but I can't remember which one I test last night). I also buying Intel AX210 to replace the Mediatek MT7921 Wifi card that ASUS love to use on all its laptop.

That device cleanup tool might be useful for my desktop, as the windows is from windows 7 i5 4670k era that I update periodically, thanks!

you could also be affected by this (win32 priority separation setting):

Win32PrioritySeparation - Understanding How the Values Impact Your Performance - XBitLabs

Changing your Win32PrioritySeparation value in the registry has become a staple in the optimization process of your PC, but unfortunately, a lot of people editing their values rely on guesswork or internet myths. Using this or that value just because someone in a Reddit thread mentioned it is...

www.xbitlabs.com

For example: A value like 22 Decimal/16 Hex – Long, Variable, High foreground boost – has all of the characteristics to increase FPS performance in game since it aggressively focuses the game you play. This is denoted by the long (minimizing context switching which is taxing on the CPU), variable quantum (can hold CPU for longer on the application), with a high foreground boost (CPU spends 3:1 time on the game favoring it heavily).

Longer quantums are commonly used in server versions of Windows but can also lead to improved FPS performance in games and higher stability in audio applications. Shorter quantums should theoretically improve responsiveness since the data is refreshed faster and is more up-to-date for your foreground application (game).

i am running decimal value of 22: variable, long, max foreground boost. be sure to also disable windows game mode and also disable foreground boost from proccess lasso options->cpu->foreground boosting (this way youll still get foreground boosting via win32 priority separation and also too many different foreground boosts running at the same time will cause issues):

you can run the device cleanup tool every 2 weeks or once a month as ghost devices will pile up over time by default. theres also a version for commandline. actually there should be a windows function for this its so essential cleaning up those devices.

and by the way heres another timer:

Network Stack (Packet Scheduler) Timer Resolution

www.overclock.net

 

[Apocalypsee]

This is yesterday latencymon

And today it running fine? Odd

DPC Latency always in the yellow no matter what. This is at idle not doing anything

My bcd is short and nothing much

 

[DaemonForce]

DPC Latency always in the yellow no matter what. This is at idle not doing anything

This is an interesting tool. No idea if I can do anything with it but I can at least measure some of my systems with it.
This Ryzen box:

The Athlon box:

Seems legit.

 

[zoulztealer]

windows booting and loading everything up (yellow), then idling (green):

idling after booting / all loaded (av + win defender, internet, process lasso, islc, nvcpl and msi afterburner running in background), seems im still able to getting lowest possible latency:

steam and browser running:

This is yesterday latencymon

And today it running fine? Odd

DPC Latency always in the yellow no matter what. This is at idle not doing anything

My bcd is short and nothing much

latency mon looking good, dpc lat not so much.

try adding:

bcdedit /set disabledynamictick true

and then make another bench with dpc late and see if you get the yellows away. right now your timer / cpu runs in laptop mode with power saving and reduced performance and bad latency.

also try running islc (intelligent stand by list cleaner) + custom timer 0.5 and process lasso in the background with bitsum highest performance or 1usmus ryzen universal power plan power plan and disable core parking with park control (core parking is disabled by default in the process lasso bitsum highest performance power plan).

1usmus Custom Power Plan for Ryzen 3000 Processors v1.1 Download

This custom power plan for AMD Ryzen 3000 "Zen 2" processors will ensure that workloads do not bounce between cores, leading to higher and more stabl

www.techpowerup.com

This is an interesting tool. No idea if I can do anything with it but I can at least measure some of my systems with it.
This Ryzen box:

The Athlon box:

Seems legit.

yes, it cant do much other than read and show results.

id recommend same for you and add

bcdedit /set disabledynamictick true

process lasso, islc + custom timer res 0.5 and highest performance windows power plan.

heres several custom planz for intel & amd and where i got mine from i run currently and am very satisfied with:

Ryzen Custom Power Plans for Windows 10/11 (Snappy...

www.overclock.net

and power plan explorer if you want to make your own one:

Windows power plan settings explorer utility

http://www.mediafire.com/file/wt37sbsejk7iepm/PowerSettingsExplorer.zip MD5 hash for zip-file: 4E5171D0F146BCD0AEF128256FC4B304 I got sick of...

forums.guru3d.com

1usmus ryzen universal would generally be a very good choice for older or other cpus. you can theoretically use amd and intel power plans vice versa.

id also like to remind you all of the msi utility v3 and the microsoft interrupt affinity tool which can help also to make the system smoother:

_IRQ_DEVICE_POLICY (wdm.h) - Windows drivers

The _IRQ_DEVICE_POLICY enumeration (wdm.h) type indicates the operating system policy used to assign interrupts from a device to different processors.

learn.microsoft.com

choice of irq policies, interesting values for high performance are 1 (numa), 3 and 5:

Windows: Line-Based vs. Message Signaled-Based Interrupts. MSI tool.

... or another attempt to improve latencies Little bit of theory: ***** From "Windows Internals" by Mark Russinovich, David A. Solomon, Alex...

forums.guru3d.com

Microsoft Interrupt Affinity Tool Download

The Microsoft Interrupt-Affinity Policy Tool (IntPolicy) is a software that allows you to "bind" or change the CPU affinity of the interrupts for a g

www.techpowerup.com

ive simply enabled everything and set it to high priority. ive never encountered any issues with that but the author warned there could be issues setting your ssd, nvme or whatever and it may not be able to boot anymore.

nvcleanstall also gives similar and further options for msi and installing the driver in the most minimal with advanced tweaks yields noticeable improvements in latency, performance and compatibility. be sure to always clean your drivers with display driver uninstaller from safe mode before updating gpu drivers:

NVCleanstall v1.17.0 Download

NVCleanstall lets you customize the NVIDIA GeForce Driver package by removing components that you don't need (or want). This not only keeps things ti

www.techpowerup.com

Display Driver Uninstaller (DDU) V18.0.8.0 Released. - Wagnardsoft Forum

and i just realized on the screen from overclokcers.at he gets like 683 fps with gtx 1080 ti and i only 452 with rtx 2080 ti and his frametimes are way more flat, etc. seems like there could be sth to this, think ill make some changes and check soon:

And today it running fine? Odd

View attachment 362125

high ndis can be cause by "bad setting" of network throttling index:

read the spoiler under the network throttling index chapter: youll need to adjust the value according to your desired bandwidth. ive set mine to 80 which should equal roughly 10mbYte/s. not sure though how this affects (max) download speed exactly but it reduces ndis dpc. you should also tweak your network adapter from device manager.

NDIS.sys DPC latency is increased

GamingPCSetup/CONTENT/RESEARCH/NETWORK at master · djdallmann/GamingPCSetup

A research and evidence based approach to optimizing your gaming PC, configuration and setup. Recommendations found in this guide are based on curated reputable technical references, and personal r...

github.com

Network

A research and evidence based approach to optimizing your gaming PC, configuration and setup. Recommendations found in this guide are based on curated reputable technical references, and personal research.

djdallmann.github.io

Gaming and mouse response BIOS optimization guide for...

www.overclock.net

ive recently created a pastebin of my best tcpip tweaks which can each be run from powershell with admin rights (though these tweaks work perfect and wont do harm, be careful, these are only for advanced users):

Set-NetAdapterRdma -Name "*" -Enabled $TrueSet-NetAdapterRsc -Name "*"-IPv4Ena - Pastebin.com

Pastebin.com is the number one paste tool since 2002. Pastebin is a website where you can store text online for a set period of time.

pastebin.com

 

[Apocalypsee]

latency mon looking good, dpc lat not so much.

try adding:

bcdedit /set disabledynamictick true

and then make another bench with dpc late and see if you get the yellows away. right now your timer / cpu runs in laptop mode with power saving and reduced performance and bad latency.

It's still yellow, AX210 card arrived today and I swap it. Still the same.

I did the disabledynamictick, win32priorityseparation to 26 hex (this makes framerate more stable in cyberpunk and in Fallout London and no crashing!), run process lasso in background. The audio stutter is still there but it occur far less now.

Further tweaking, this is before

I did a few more, disable C-states and Intel Dynamic Tuning Technology in BIOS but latency is still there, wdf01000.sys latency now so it's not related to power saving. I found a tweak and that fixed the latency problem once an for all. See below. The Processor idle demote and promote threshold is the culprit. To see this options download PowerSettingsExplorer and untick both of this options in there, it should appear in Windows power options after this. I'm going to play games now to see if the audio drop is gone...

 

[zoulztealer]

I did a few more, disable C-states and Intel Dynamic Tuning Technology in BIOS but latency is still there, wdf01000.sys latency now so it's not related to power saving. I found a tweak and that fixed the latency problem once an for all. See below. The Processor idle demote and promote threshold is the culprit. To see this options download PowerSettingsExplorer and untick both of this options in there, it should appear in Windows power options after this. I'm going to play games now to see if the audio drop is gone...

View attachment 362362

now this bench looks way better than all the others! did it change dpc latency checker too or only in latency mon?

but be sure that you do not disable the function in the bios that clocks your cpu up: i think for intel its / was called eist / speed step or sth. not sure what dynamic tuning technology does exactly but if you disabled it and your ram, cpu, gpu and everything else still reach their max clocks it shouldnt be an issue. apart from that you can disable every other "energy saving" function.

i dont remember what setting it was exactly but for ryzen cpus even you disable every available power saving function the cpu will still use the option to put the cores asleep or else it cant clock at all. this will probably the same for some functions from intel and others too. so you dont need to worry that your laptop will use more power. the difference is only marginal but the performance will me much better.

for my cpu / mb i have enabled basically all power saving functions there is with a very few exceptions and everything works fast and efficient to my surprise. i was afraid i will have to disable all power saving functions in bios too and hpet as well but to my surprise it wasnt necessary yet i still believe performance and latency would even better if i did. i have 2 other ryzen (1700 and 3600) here that all run in the same manner with same tweaks (not precisely all the same but very similar) perfectly in all aspects: very power saving for office / desktop and efficient and performant for gaming, multimedia stuff and multitasking.

remember that you still have several more options to go to hopefully and eventually eliminate the audio crackling if its still there.

your screenshot also doesnt show the "optimal" power plan: use either bitsum highest performance or 1usmus ryzen universal or other ones that are optimized for intel / your cpu. you can install the powerplans with command line powercfg.exe. i would also disable fast boot and hibernation if you dont need / use it specifically and theres also nice tools like o&o shutup win10, and w10 privacy that can help reduce background cpu load and which i both used extensively to disable stuff i dont use in windows.

be aware that if you update windows or drivers it may be you need to re-align some of settings and tweaks you did; for example every time you update gpu drivers you will need to reapply tweaks from nvcleanstall or msiutilv3 and windows updates may enable shady or unneeded windows services or settings again.

How to use powercfg to control power settings on Windows 10

Yes, you can control various power settings with Command Prompt, and in this guide, we'll show you the steps to use the powercfg tool to complete many power tasks on Windows 10.

www.windowscentral.com

ManniX_Custom_Power_Plans_V2.zip

drive.google.com

if you run process lasso it should also automatically engage bitsum highest performance power plan and / or switch power plans (which you can turn off from options). bitsum highest performance power plan has already core parking disabled by default. process lasso also fixes performance issues of intel cpus with e+p core configuration in games and it also improves performance / core distribution on ryzen.

if you use another power plan core parking is likely enabled: you can check and configure core parking of your power plan with park control or power plan explorer of course, too.

as you pointed out there are dozens of important settings to the power plan that can be viewed and modified from the power plan explorer and therefor it would be best to use an already customized plan from someone like 1usmus or mannix or if you create your own one after you tested and understood each setting as these settings do have a definitive impact of behavior and performance

 

[Apocalypsee]

After installing the power plan, I chose the Intel Balanced Snappy. Here is the dpc latency now. I read that dpc latency checker is not meant to run on newer Windows anyway.

FURTHER EDIT:
Only a few power profile is available idk what happen to tohers that I install

It's almost there, only random spikes from ACPI.SYS that occur sometimes, and when it happen it have audio drop, whether I'm gaming or watching videos.

 

[zoulztealer]

After installing the power plan, I chose the Intel Balanced Snappy. Here is the dpc latency now. I read that dpc latency checker is not meant to run on newer Windows anyway.

FURTHER EDIT:
Only a few power profile is available idk what happen to tohers that I install

It's almost there, only random spikes from ACPI.SYS that occur sometimes, and when it happen it have audio drop, whether I'm gaming or watching videos.

very nice! if you have realteak audio, clean uninstall the drivers with display driver uninstaller from windows safe mode, it has a special option for cleaning realtek audio drivers.

then check again hows your latency and audio without the drivers; for most audio cards you dont need an extra driver, you can just use the windows default one which has way better performance.

also run autoruns from ms sysinternals and disable everything you dont need in the background from startup, services and drivers:

Autoruns - Sysinternals

Lassen Sie sich anzeigen, welche Programme so konfiguriert sind, dass sie automatisch gestartet werden, wenn Ihr System bootet und Sie sich anmelden.

learn.microsoft.com

https://download.sysinternals.com/files/Autoruns.zip

the acpi.sys high dpc latency seems to be a common and widespread issue among laptops.

ms forums: High DPC latency from acpi.sys causing audio clicks and pops

https://www.reddit.com/r/windows/comments/1670ej3

 

[:D:D]

Wait till you take into consideration not all HPET hw implementations are equal.

Or Windows builds.

Surprising Observation! HPET On vs. Off - Page 3 - Blur Busters Forums

View attachment 361853

Using above settings X99

HPET can be used as a timer and/or counter. Usually RTC is used as a timer if HPET is disabled (BIOS or register) and platformtick is requested. It's unusual that it runs at a timer resolution of ~0.4882ms and presumably Windows counts the number of interrupts before doing something other than just acknowledging, so 32 interrupts for 15.6ms for example. This also means min and max resolution ca be set lower or higher than OS limits.

Example at 125ms

Also no noticeable pops, cracks or whistles with audio with that setting although testing was very limited.

Seems disabledynamictick on this system causes timer interrupts to only only occur on CPU 0 instead of any CPU when LAPIC is used.