GSYNC Breaks In-Game VSYNC

Discussion in 'Videocards - NVIDIA GeForce Drivers Section' started by PhantomGamers, May 22, 2017.

  1. khanmein

    khanmein Ancient Guru

    Messages:
    1,648
    Likes Received:
    71
    GPU:
    EVGA GTX 1070 SC
    @RealNC Thanks & greatly appreciated. I don't own any G-Sync.

    FYI, vsynctester.com value is variable. Can I use MPC-HC + MadVR (Ctrl +J) = 59.9491 Hz?

    Which one exactly -0.005, -0.006 or -0.007?
     
    Last edited: May 24, 2017
  2. RealNC

    RealNC Ancient Guru

    Messages:
    3,152
    Likes Received:
    1,379
    GPU:
    EVGA GTX 980 Ti FTW
    You only need the first three decimal digits. Also, make sure you use a browser that works well (Chrome and Firefox should work) and ALSO make sure you have Aero enabled (if you're on Windows 7.)

    MPC-HC+MadVR AFAIK don't test your refresh rate. They just read the timings. The monitor might actually not run at exactly that. vsynctester.com should be able to get your actual refresh. Again, you only need the first three decimals (59.xxx).
     
  3. RealNC

    RealNC Ancient Guru

    Messages:
    3,152
    Likes Received:
    1,379
    GPU:
    EVGA GTX 980 Ti FTW
    This only applies to the nvidia limiter. No reason to set fast sync if you use RTSS or in-game capping. Although if you do nonetheless, it will behave the same as if you had selected vsync instead.

    Unfortunately, I'm not aware of any tests on frame time pacing with G-Sync. Last time I looked, it seemed no one has equipment for FCAT + G-Sync. And that's very expensive equipment to begin with, mind you. Guru3D does have an FCAT setup AFAIK, but probably not for G-Sync, only for fixed refresh. I don't even know if it's even possible to FCAT G-Sync. A setup for this might not even exist at all.

    Oh, if that's what you mean, the new setting fixes that. It now limits to exactly what you've set it to.
     
  4. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,371
    Likes Received:
    906
    GPU:
    1080Ti H20
    I meant both, I should have been more specific.

    I'll monitor battlenonsense channel and that thread as well for more info.
     

  5. khanmein

    khanmein Ancient Guru

    Messages:
    1,648
    Likes Received:
    71
    GPU:
    EVGA GTX 1070 SC
    By the way, I'm using W10 & I tested on vsynctester/mpc gave me the same results.

    One last question, e.g. Dota 2 I turned V-Sync Off & max frames per second allowed set 240 FPS (both under game setting) + Fast Sync enabled (NvCpl) + RTSS capped 59.942/60, during game play I get around ~120 FPS (MSI: AB + RTSS off) & ~60 FPS (MSI: AB + RTSS on) "weird"

    Both scenarios don't have any screen tearing & I can't even differentiate what's the difference since my monitor is just 59//60 Hz only.
     
  6. RealNC

    RealNC Ancient Guru

    Messages:
    3,152
    Likes Received:
    1,379
    GPU:
    EVGA GTX 980 Ti FTW
    That's not weird. If you don't cap, fast sync is going to try and lock to a multiple of your refresh. That means you get 120FPS, 180FPS, 240FPS. It will usually jump between these exact multiples.

    If you cap to 59.942, RTSS usually reports 60. There's a setting in the latest RTSS betas that allows you to turn on fractional FPS display and disable FPS averaging.

    To use the "low latency vsync" trick (which is what you just did, capping to ~0.007FPS below refresh) I recommend just using vsync, not fast sync. That's simply because I don't know what internal quirks the driver might be doing when using fast sync.

    With that being said, an input lag difference is very difficult to tell in DotA. You will be able to tell in other games that have a mouse-controlled 1st or 3rd person camera. Playing Doom 4 for example with this trick makes the game feel "instant" with almost no "floating mouse" effect.
     
    Last edited: May 24, 2017
  7. khanmein

    khanmein Ancient Guru

    Messages:
    1,648
    Likes Received:
    71
    GPU:
    EVGA GTX 1070 SC
    Gracias, I owned Doom 4. I'll try & see.
     
  8. Mda400

    Mda400 Master Guru

    Messages:
    851
    Likes Received:
    90
    GPU:
    1660 - 2130/10ghz
    Show me a video and then i'll eat crow. I can make bar graphs and charts all day.

    Don't show me this, I want to see something like this

    "well why don't you go make it yourself?" because i'm not rich and I'm not buying something I won't use later on.
     
    Last edited: May 24, 2017
  9. RealNC

    RealNC Ancient Guru

    Messages:
    3,152
    Likes Received:
    1,379
    GPU:
    EVGA GTX 980 Ti FTW
    Is there an ignore list so I don't get to see his posts?

    If you don't want to use a piece of software, that's completely OK. No one is forcing you to use anything you don't want. But don't try to ruin other people's conversations. Can you do that?

    Yes, all tests are fake. I admit it. Neither we nor Battle(non)sense actually performed any of the tests. I ADMIT IT.

    Can you now stop interfering?
     
    Last edited: May 24, 2017
  10. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,371
    Likes Received:
    906
    GPU:
    1080Ti H20
    If you think Battle(non)sense and blurbusters are giving fake graphs you are to be blunt, an idiot.

    They both did the same setup(a better setup infact) that Linus did.

    My advice, stay out of these threads and keep to yourself.

    Yes there is an ignore list.
    Probably a good idea.

    Click on user name > View public profile. under their name you will see users list > Add to Ignore list.
     

  11. Mda400

    Mda400 Master Guru

    Messages:
    851
    Likes Received:
    90
    GPU:
    1660 - 2130/10ghz
    Is it a problem that I question the validity (was I EVER off-topic)? This is suppose to be a discussion. But you want it to be a presentation and the nay-sayers to sit down.


    Yes, much better to ignore than think about the possibilities... :bang:

    If i listened to your advice, i would never have found how to make something you assume is worse, better.
     
    Last edited: Jun 30, 2017
  12. jorimt

    jorimt Active Member

    Messages:
    56
    Likes Received:
    25
    GPU:
    EVGA 1080 Ti FTW3
    As the author of the original post, I agree about the information being unfinished overall. I'm flattered everyone refers to it as an "article," but it was only ever intended to be a forum post.

    I have however, been working on the article version of "Input Latency" for the past month, and am still going. Input latency tests are extremely time consuming, since every single frame has to be counted, one at a time, ad nauseam. There are also a seemingly infinite amount of combinations to test, so once you think you've come to a conclusion, you remember another that could act as a cross-reference/control, and it never ends; there's a reason few have attempted it.

    As for CS:GO's "Multicore Rendering" settings, yup, that discrepancy was unfortunate. A main reason why I decided to use Overwatch as the main test game for my upcoming article, seeing as the DX9 CS:GO engine is outdated, and is not a native multi-core/multi-threaded performer.

    You are fully entitled to your opinion, and and are also entitled to question the validity of the results. I think the main issue here, is you haven't made yourself terribly clear. Do you mean to say you think every test other than Linus' tests are invalid, and both Battle(non)sense and I are simply coming up with numbers and punching them into graphs?

    If so, I know better, but I can't convince you otherwise, and I won't try.

    For those who don't totally question our methods (which will be explained and pictured in extreme detail in my upcoming article), I hope to put a few questions to rest soon. But no, due to all the possible system configurations, nothing will ever be "for certain."
     
    Last edited: May 25, 2017
  13. Mda400

    Mda400 Master Guru

    Messages:
    851
    Likes Received:
    90
    GPU:
    1660 - 2130/10ghz
    It is well-written though, and you have laid out a nice picture of how G-sync technically behaves, which before reading I did not know as I assumed G-sync was replacement for V-Sync (that when it was enabled, G-sync would interact with the driver and set its own 'hardware' frame limit at max refresh rate to keep from tearing when going above essentially doing away with V-sync's double buffer). But I see that the top end relies on V-sync unless you force fast-sync or frame-limit through other means.

    In summary I thought it was an adaptive hardware frame limiter that communicated with the display on what refresh rate to run at every hz and at the same time, wouldn't allow the framerate to exceed monitor refresh rate. But since it works in driver and not relying on application support, the renderer will still get congested when it reaches the refresh rate.

    At least you know there are many variables that can affect these results, but documentation of the tech itself is very handy.

    It's just a personal preference. The LED and mouse test with a highspeed camera is the most accurate I know of since the LED is an affirmation of when the mouse click is sent and is going through the video chain to happen in the application. I'm not very good with soldering, else i would make the same device. There are others that have done this method with non-Gsync/FreeSync monitors, but as far as I know Linus did it first for G-Sync/FreeSync.
    .
    But for my own assurance (and I don't expect anyone to care about mine), to be able to see the LED test with RTSS vs Inspector in its most current versions then I would take that as enough evidence. I could tell there was more delay induced once Nvidia revised frame limiter to v2 with the addition of fast sync in its drivers (there was even a difference between 1.9.7.6 and 2.1.3.x v1 limits). Luckily Inspector just needed a revision to force v2 behavior off (yes RealNC, it is better with 2.1.3.6).
     
    Last edited: May 26, 2017
  14. jorimt

    jorimt Active Member

    Messages:
    56
    Likes Received:
    25
    GPU:
    EVGA 1080 Ti FTW3
    I'm not sure what you've thought up until now, but that's the exact method both I and Battle(non)sense are using.

    As for Linus being the "first," he was merely borrowing the method from Mark Rejhon, who was the first to use it in his 2014 "Preview of Nvidia G-SYNC" article on blurbusters.com. Linus literally says it near the 1:51 mark in the video you've been referencing (forgive my inability to post direct links; I haven't made enough posts to use links ;p).

    Also Linus had no proper control to compare the results against, and I still don't even know what they mean.

    You may take that back if you knew how much work it was. A single video with a ten sample count takes me up to five minutes to inspect. Times that by four videos for each scenario, and counting through only two scenarios would take you up to forty minutes with my method.
     
    Last edited: May 26, 2017
  15. Mda400

    Mda400 Master Guru

    Messages:
    851
    Likes Received:
    90
    GPU:
    1660 - 2130/10ghz

    If you use this test, then why not post a video on it? Sorry for my ignorance, but like i said i've only seen video evidence by Linus on this.

    Just something like this is all i've been looking for between frame limiters.
     
    Last edited: May 26, 2017

  16. jorimt

    jorimt Active Member

    Messages:
    56
    Likes Received:
    25
    GPU:
    EVGA 1080 Ti FTW3
    Honestly, I didn't see the need in my original post, seeing as I provided a link to Mark's testing method, and there was precedence for it. I also wasn't aware there were those like you who would get the impression that we were simply coming up with numbers randomly and popping them into graphs.

    I have, however, remedied this for my upcoming article, which has a very large section dedicated to explaining my test methodology, as well as a looping video showing one of the test videos in action. I'll also have my original Excel spreadsheet available for closer inspection for those interested.
     
  17. Mda400

    Mda400 Master Guru

    Messages:
    851
    Likes Received:
    90
    GPU:
    1660 - 2130/10ghz
    Sorry to be blunt but, welcome to the internet? I hope that many are reserved before acknowledging information.

    It would be helpful and appreciated. I know you don't owe me anything, but if you are trying to prove a point, only thing that will drive it home for me would be example above since what we are discussing here is related to video.

    Yes anything on here is to be taken with a grain of salt, i know that and have said it at times when i reply. But it never hurts to start the discussion.
     
    Last edited: May 26, 2017
  18. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,371
    Likes Received:
    906
    GPU:
    1080Ti H20
    So anyone who doesn't have a video for you to watch is instantly discredited and you assume they pull numbers out of their ass?

    Do you realize the amount of work it would take to video every single test?
    Jeeze you are dense.

    Where's your video proof that NV inspector is better in every way compared to RTSS?

    Oh wait you don't have one therefore you are pulling numbers out of your ass..

    Get the point yet?
     
    Last edited: May 26, 2017
  19. jorimt

    jorimt Active Member

    Messages:
    56
    Likes Received:
    25
    GPU:
    EVGA 1080 Ti FTW3
    As I laid out in my previous reply, I can only offer an example of the process. Showing a video of every single result, especially in article format, would be entirely impractical as I have hundreds of videos totaling in the multiple GBs. I'm also aware the average reader won't want to watch multiple 5 minute videos in slow motion, and would rather view the results in the finished chart/number form.

    You're idea of proof may differ from mine, and mine may not meet your standards ultimately, but, you're right, it is the internet, and you can't please everyone, so mine will have to suffice for the "everyone else" willing to accept what I offer.

    If you want the level of proof you are seeking, the only alternative is to modify your own mouse, learn the methodology, and perform the tests yourself.
     
  20. RealNC

    RealNC Ancient Guru

    Messages:
    3,152
    Likes Received:
    1,379
    GPU:
    EVGA GTX 980 Ti FTW
    I don't see that guy complaining when reviews show up on the front page of Guru3D with only the benchmark results and no "video proof."

    Nobody posts videos when they publish a paper with results in scientific journals.

    And even if you'd post videos: The moon landings are fake and the footage is fabricated.

    Give me a break.
     

Share This Page