Another look at HPET High Precision Event Timer

Discussion in 'Videocards - NVIDIA GeForce Drivers Section' started by Bukkake, Sep 18, 2012.

  1. Raiga

    Raiga Maha Guru

    Messages:
    1,099
    Likes Received:
    0
    GPU:
    GPU
    Best leave HPET as it is..don't mess with it.

    Reason being, that all the current hardware and drivers are; configured and tested with default stock windows settings by the manufacturer.

    Some sources I currently checked on the net

    http://stackoverflow.com/questions/...ndows-system-time-with-millisecond-resolution

    http://msdn.microsoft.com/en-us/magazine/cc163996.aspx#S8

    It mentions that.. Windows 7 has a default 15ms timer resolution (minimum)...

    But on my system its 0.917ms maximum value (close to 1ms) anyways..
    Translates to 0.891-0.905ms kernel latency timer (via latencymon)

    changing the timer resolution to 0.466ms (close to 0.5ms) via the tool
    Translates to 0.406-0.421ms kernel latency timer (via latencymon)

    Funny thing is, I haven't even enabled HPET in Bios or in Windows 7 via the CMD command (lol) to change timer resolution.
     
  2. warlord

    warlord Master Guru

    Messages:
    691
    Likes Received:
    0
    GPU:
    R9 390X MSI(1080/1500)
    so...the reasonable gamer and power user knows these facts are the only way:

    1)HPET on bios -> on,
    2)HPET on windows -> Go to CMD and type bcdedit /set useplatformclock true
    3)Check with WinTimerTester that frequency is 14.31818

    14.31818 frequency is the only way to prove hpet is both on in os and bios

    4)Check with DPC latency checker tool

    5)download timer resolution tool (free) and run it as administrator

    6a)"choose the "maximum" which is 0.5 milliseconds (default system has 1ms)

    6b)it should be kept open

    7)then check with DPC latency checker tool, latency should be cut exactly in half

    CONGRATULATIONS YOU NOW HAVE THE SMOOTHEST POSSIBLE EXPERIENCE


    this post should be sticky somehow, because it works 100% for everybody who made all the steps above exactly as mentioned.
     
    Last edited: Jan 7, 2013
  3. fellix

    fellix Member Guru

    Messages:
    144
    Likes Received:
    0
    GPU:
    EVGA GeForce GTX980 Ti SC
    Does the version for XP work on Win8? V2.0 is behind pay wall.
     
  4. warlord

    warlord Master Guru

    Messages:
    691
    Likes Received:
    0
    GPU:
    R9 390X MSI(1080/1500)
    it works mate, we all use that...it's just a marketing trick by author i believe :)
     

  5. Penal Stingray

    Penal Stingray Banned

    Messages:
    957
    Likes Received:
    0
    GPU:
    GTX 680 Tri-Sli-S27A950
    I can vouch for this that latency timer tool setting to .05ms resolution elimantes the hiccups I get when turning hpet on windows. make sure ur don't close the tool keep it running while playing game closing it will revert the resolution back to 1ms. also one thing it may not work for some configuration but it worked perfect on mine
     
  6. Xtreme512

    Xtreme512 Master Guru

    Messages:
    591
    Likes Received:
    0
    GPU:
    GTX1060 6GB
    [​IMG]

    is it Ok ? hpet is enbaled in bios by stock and thats all.. ı didnt touch anytihng else in windows.
     
  7. LucasHale

    LucasHale New Member

    Messages:
    2
    Likes Received:
    0
    GPU:
    Nvidia
    Hi,

    I'm the author of the program (TimerResolution), I get quite a lot of traffic from Guru3D so just want to say thanks for your support!

    The main difference with Version 2 is that it has a command line so that it can easily be made to run at start up and set the timer resolution to the maximum value. It is also supported on the new versions of windows but the old version does seem to work for most people.

    But as always, if you find the tool useful and appreciate it then buying the paid version is a nice thing to do :)

    Feel free to email me any questions about the tool.

    Regards,
    Lucas
     
    Last edited: Jan 8, 2013
  8. Penal Stingray

    Penal Stingray Banned

    Messages:
    957
    Likes Received:
    0
    GPU:
    GTX 680 Tri-Sli-S27A950
    Lucas Hale plz check ur PM thank you.
     
  9. warlord

    warlord Master Guru

    Messages:
    691
    Likes Received:
    0
    GPU:
    R9 390X MSI(1080/1500)
    i never meant to say these 2 versions are the same and i am glad to know that there are differences :D

    i will try the 2.0 version as soon as i find out how to do this, the only buys i do online is from steam :)

    i see AUD, but in my country i use EUR, how can i do this, please pm.
     
  10. Penal Stingray

    Penal Stingray Banned

    Messages:
    957
    Likes Received:
    0
    GPU:
    GTX 680 Tri-Sli-S27A950
    Dude ur busted! lol He caught you man real good! lol I bet you didn't see that one coming lol heck I was surprise he just popped out of nowhere lol!
     

  11. tsunami231

    tsunami231 Ancient Guru

    Messages:
    7,666
    Likes Received:
    15
    GPU:
    EVGA 660gtx sig2
    hmm 10$ just to have the ability to hide and autorun the program seem bit steep imo, will see if it even does anything noticable first

    So far only thing i noticed it did was Lower latency if I used dpclat100us+ or 10us with timer, mean while Latency monitor shows same exact latency I always had with it on or off which is 10us
     
  12. tsunami231

    tsunami231 Ancient Guru

    Messages:
    7,666
    Likes Received:
    15
    GPU:
    EVGA 660gtx sig2
    after more testing. DPClat & Latencymon show completely different numbers Latecnymon show average of current 10us under idle and load at most time with Highest spikes coming routine us from nividia drivers.

    Mean while dpclat show 100us under idle and about 500us under load if timer is set to max it indeed does drop dpclat idle to 10us and load to about 100us.

    Timer indeed does something according to DPClat and is much more obvious, Latecnymon dont seem to do much or atlest its alot less noticeable. Overall I like the program, not fond of fact I need to pay for simple hide function and auto start though. To bad the os cant be set to just use .500 at all times permanently

    Which program is more accurate is the question. And I personally think Latecnymon is more accurate.


    Think good way to test it would be some one with audio poping, with HPet on bios&windows runs the timer on max and see if the audio popping goes away

    Great program none the less sidenote i know about this program for years but never both with it cause i never put much stock in latency the last year or 2.
     
    Last edited: Jan 8, 2013
  13. LucasHale

    LucasHale New Member

    Messages:
    2
    Likes Received:
    0
    GPU:
    Nvidia
    I think the reason the Microsoft wont leave the timer at 0.5ms is that it is a global timer and has an impact on all sorts of things - one of the negative side effects of this is that power consumption goes up which means shorter battery life on laptops etc.
    Basically MS is relying on the people who write the software to know what they are doing and if they need a higher resolution timer then they should ask for it - based on how many different games and audio programs seem to improve when used in conjunction with TimerResolution it seems that MS may not have clearly explained this feature to developers.

    To cover of the question about charging for Version 2, Version 1 was written back in the Win XP days when the max resolution was ~1ms but I coded it to be flexible (which is why it mostly works on newer version of windows) but it didn't have a command line interface and actually wasn't meant to be used for much except for as a light weight way to get better timers and for "soft real-time". Then gamers found it and my simple little website started hitting its monthly bandwidth limit!
    Then a Counter Strike clan contacted me and asked if I could modify the program to have the command line so it would start automatically because they wanted to run it on their servers as a service and they were willing to pay. They paid $100 and got an unlimited license to use on as many servers as they liked (they were a big clan and ran many servers) - they were happy and I used the $100 to move my hosting to something bigger.

    I didn't think it was right to charge those guys and then give it away to everyone else so I decided to make both version available, the free version and for those wanting the addition features a paid version. At the time the Aussie dollar was about USD$0.75 so for most people around the world it would be a < $10 purchase (this was before the App Store and $0.99 downloads :) ) which seemed reasonable. I decided to leave the free version alone with out adding nag screens or crippling any of the features.

    Back when I only had 1 version I had a donate button on the site, I received a grand total of one donation out of 1,000s of downloads. I'm not doing it to make money but I do want to break even at the end of each year.

    Many people from Guru3D have purchased Version 2 and again I just want to say thanks for the support and I'm glad you find the tool useful!
     
  14. hulawafu77

    hulawafu77 Member Guru

    Messages:
    191
    Likes Received:
    0
    GPU:
    7970M
    Don't know if it works or not, but I'd pay for it if it was $5. $10 seems bit steep to me. I don't even pay $10 for my games, most of the Steam/Amazon/GreenMangaming etc bundles I got, each game on average about $5.

    Anyway, I'll keep tab on it. I have HPET enabled, verified with WinTimer, new about this for a fix with overclocking with SetFSB.
     
  15. mbk1969

    mbk1969 Ancient Guru

    Messages:
    3,571
    Likes Received:
    13
    GPU:
    GeForce GTX 970
    Setting the "useplatformclock=true" you force OS to interact with so called platform chip - either LAPIC or HPET.
    When I tested "useplatformclock=true" (no metter with HPET or LAPIC) LatencyMon`s stats showed me that ISR (or DPC or both - I don`t remember now) count increased nearly twice. And there was no difference in routines longest execution time. So I choose less interruptions (without disabling HPET - "useplatformclock=false" - in case if some program needs it).

    QPF on my rig is 3,5MHz. 1/3,5MHz = 0,0029ms - pretty little time quantum imo. I don`t know such process that need less time quantum for a timeout measures.

    PS Forgot to mention that modern processors features Invariant TSC which not affected by C-states, P-states and multicore desync. Check CPUID...
     
    Last edited: Jan 17, 2013

  16. Xtreme512

    Xtreme512 Master Guru

    Messages:
    591
    Likes Received:
    0
    GPU:
    GTX1060 6GB
    guess what my hpet was enbaled in biod and windows by default.. I proved it with 14.3 Mhz thing in that tool.

    and for crysis 3 I tested 0.5 timer tool
    made a little difference in smoothness..

    this tool gives more advantage for users that have weak cpu.. no big gains for me like ivy 4.5 GHz. ;)

    But again its a good technique ;)
     
  17. maco07

    maco07 Active Member

    Messages:
    96
    Likes Received:
    0
    GPU:
    7970 3GB Boost
    I don't think 2) is right. Take a look at this:

     
  18. Major Melchett

    Major Melchett Ancient Guru

    Messages:
    2,456
    Likes Received:
    0
    GPU:
    R9 280X @ 1100/1500
    Really strange, I tested the timer tool last night and it was reporting my default at 1.000ms, things did improve a little when I set it to 0.500ms. However today when I started the tool my default is now 10.000ms.

    Weird.
     
    Last edited: Feb 26, 2013
  19. maco07

    maco07 Active Member

    Messages:
    96
    Likes Received:
    0
    GPU:
    7970 3GB Boost
    read previous pages: You need to change it every time you start windows and leave it open. Or pay 10 bucks for command line/hide version. A little expensive taking in count that I paid 40 for my Win8 Pro.
     
  20. Major Melchett

    Major Melchett Ancient Guru

    Messages:
    2,456
    Likes Received:
    0
    GPU:
    R9 280X @ 1100/1500
    That's not what I was talking about, I was stating that it's weird that last night it reported a default timer of 1.000ms and today it reports a default of 10.000ms (this is before even clicking the maximum button).
     

Share This Page