Surprising Observation! HPET On vs. Off

[imprecise]

Recently I dived deep into the whole Windows Tweaker scene. I was going trough alot of Forums with People claiming that there are Tweaks that can have a big impact on the performance.

Disabling HPET is probably the most popular Tweak.

This is the current thing I've been testing. I have used it in the past but not recently until a few days ago. Even after doing BIOS, Windows, and other tweaks I found this HPET setting to make an impact. With HPET off the mouse input feels very responsive and consistent. I notice in CSGO my framerate goes over the 400 cap now and hits 404 often, and there are other odd desync moments and lag that still happen and show up in net_graph, but the input feeling actually remains consistent. Even after an obvious lag spike, the responsive feeling no longer vanishes into numb unresponsive gameplay.

Before optimization, movement feeling is very loose and inconsistent. Like a worn out controller that you have to grip extra strong to overpower the lack of responsive feedback. Even without obvious on-screen lag, there are moments where the gameplay just feels completely unresponsive and numb where the only thing controlling the outcome is randomness. Trying to time the click-lag while balancing poor movement ends up causing a lot of unforced errors.

After optimization, movement feels much more consistent and if there is a moment of lag, the responsiveness recovers instantly. I do think HPET has a lot to do with this particular scenario, as it's like the 10th tweak I've configured on this system and I can still tell a difference in movement feeling. I read that Windows 11 doesn't even use HPET, so having it turned on could be causing timing delays and errors from trying to run multiple timing systems simultaneously.

Test 1 - With HPET on, lower dwm priority, increase wininit priority: big improvements in responsiveness, randomness dropped from ~20% to ~5%. Another big factor here was turning off parked cores.
Test 2 - disable HPET, responsiveness is great and reliable. Seems like memory timings might need a tweak to deal with laggy desync spikes, but responsiveness is great.
Test 3 - adjust dwm and wininit priorities to default with HPET off. Can't tell a difference, maybe 5% increase in randomness(?)
Test 4 - back to dwm and wininit priorities optimized, HPET off. This feels very consistent and reliable and I'm not too interested in eliminating dwm from the equation, as full-screen optimization(dwm.exe) made a difference in older versions of Windows as well. The game does some desync lagging but the important thing is the responsiveness doesn't change and seems to recover instantly instead of going into numb randomness for 1-3 seconds at the most critical times.

--------
Update to dwm and wininit... I've been looking to remove these adjustments from the guide and may have found the underlying cause: HPET. Currently testing HPET off and dwm, wininit back to default priorities, seems promising so far. I will be looking to update the guide with some more info, but here's a nice video about HPET:
https://www.youtube.com/watch?v=TF-jpouz7b8
and a long forum discussion:
https://forums.guru3d.com/threads/anoth ... er.368604/
and an article about HPET bug decreasing CPU performance:
https://www.overclockers.at/articles/th ... at-it-isnt
connection between HPET and synchronization drifting apart, description of desync in CSGO:
https://www.reddit.com/r/GlobalOffensiv ... er_out_of/
-----

[Sebichek45]

Recently I dived deep into the whole Windows Tweaker scene. I was going trough alot of Forums with People claiming that there are Tweaks that can have a big impact on the performance.

Disabling HPET is probably the most popular Tweak.

This is the current thing I've been testing. I have used it in the past but not recently until a few days ago. Even after doing BIOS, Windows, and other tweaks I found this HPET setting to make an impact. With HPET off the mouse input feels very responsive and consistent. I notice in CSGO my framerate goes over the 400 cap now and hits 404 often, and there are other odd desync moments and lag that still happen and show up in net_graph, but the input feeling actually remains consistent. Even after an obvious lag spike, the responsive feeling no longer vanishes into numb unresponsive gameplay.

Before optimization, movement feeling is very loose and inconsistent. Like a worn out controller that you have to grip extra strong to overpower the lack of responsive feedback. Even without obvious on-screen lag, there are moments where the gameplay just feels completely unresponsive and numb where the only thing controlling the outcome is randomness. Trying to time the click-lag while balancing poor movement ends up causing a lot of unforced errors.

After optimization, movement feels much more consistent and if there is a moment of lag, the responsiveness recovers instantly. I do think HPET has a lot to do with this particular scenario, as it's like the 10th tweak I've configured on this system and I can still tell a difference in movement feeling. I read that Windows 11 doesn't even use HPET, so having it turned on could be causing timing delays and errors from trying to run multiple timing systems simultaneously.

Test 1 - With HPET on, lower dwm priority, increase wininit priority: big improvements in responsiveness, randomness dropped from ~20% to ~5%. Another big factor here was turning off parked cores.
Test 2 - disable HPET, responsiveness is great and reliable. Seems like memory timings might need a tweak to deal with laggy desync spikes, but responsiveness is great.
Test 3 - adjust dwm and wininit priorities to default with HPET off. Can't tell a difference, maybe 5% increase in randomness(?)
Test 4 - back to dwm and wininit priorities optimized, HPET off. This feels very consistent and reliable and I'm not too interested in eliminating dwm from the equation, as full-screen optimization(dwm.exe) made a difference in older versions of Windows as well. The game does some desync lagging but the important thing is the responsiveness doesn't change and seems to recover instantly instead of going into numb randomness for 1-3 seconds at the most critical times.

--------
Update to dwm and wininit... I've been looking to remove these adjustments from the guide and may have found the underlying cause: HPET. Currently testing HPET off and dwm, wininit back to default priorities, seems promising so far. I will be looking to update the guide with some more info, but here's a nice video about HPET:
https://www.youtube.com/watch?v=TF-jpouz7b8
and a long forum discussion:
https://forums.guru3d.com/threads/anoth ... er.368604/
and an article about HPET bug decreasing CPU performance:
https://www.overclockers.at/articles/th ... at-it-isnt
connection between HPET and synchronization drifting apart, description of desync in CSGO:
https://www.reddit.com/r/GlobalOffensiv ... er_out_of/
-----

So you feel like dwm (normal) and winnit (real-time) with hpet off is the best for you? How fid you disable hpet? Just the deletevalue + disable in device manager or useplatformclock no + disabled

[imprecise]

So you feel like dwm (normal) and winnit (real-time) with hpet off is the best for you? How fid you disable hpet? Just the deletevalue + disable in device manager or useplatformclock no + disabled

Yes, I suspect HPET being on causes an additional buffer to be added to all inputs. The buffer seems random too, like maybe sometimes it's 20ms and other times it's 3-5ms. I disabled the HPET device in Device Manager, which I think was sufficient, but I also ran bcdedit /set disabledynamictick yes

dwm has always been problematic and with HPET on it did make a big noticeable difference turning down the priority. With HPET off though, I couldn't really tell if performance was degraded when I set dwm back up to default. wininit is just keeping the PC more responsive I think, as the "real-time" priority never gets applied, but process lasso does keep trying to apply it. I would definitely try turning off HPET over tweaking dwm/wininit first. I want to get away from recommending tweaking dwm/wininit, so please let me know if you test it and find anything.

In my testing on Win11 AMD 5950x I think the most important variables to turn off for better performance are HPET and core parking. Before testing the HPET setting, I would have said core parking and application priorities are most important. After testing HPET, maybe don't even need app priority changes. For me, these 2 settings made the biggest impact (~20%), and I have about 8 other optimizations listed in my guide which all make a smaller noticeable difference (~5% each).

[masneb]

There are all sorts of takes on this since time began, some people thinking its better on, others off, others with weird variations in addition to that (such as tsyn). Some people say that the HPET tweak works better on older hardware, sometimes it depends on Intel instead of AMD too. Some games also like it more/less then others as well. It's really a crap shoot. The age of the posts matters a lot as they're on different generations of hardware that have changed over the years (or even decade at this point).

I don't think there is any one take on this. I wouldn't say you shouldn't tweak it either, rather you should just try and and remember to revert it as most people don't when they mess with this stuff. I currently have all my stuff on default settings with disabledynamictick yes.

[rvteen]

disabling hpet is needed by valorant as stated by their employee on reddit to get full potential as valorant uses tsc timer instead of hpet,so it's not snake oil for valorant
btw how to get tsc timer?any command beside useplatformclock no

[SachsenPowl]

at these linked video (https://www.youtube.com/watch?v=TF-jpouz7b8&t=280s)
some user 1 year ago posted in the comments that disabling HPET give godlike mouse/input lag feeling but really bad hitreg issues/desync. Another one agrees.
I am not at home to test it on my own but is this true (Valorant/CSGO)?

[0ka]

disabling hpet is needed by valorant as stated by their employee on reddit

link? can't the developer control which timer it will use?