pneu wrote:
I hate to be a pain but are you able to retest with these RTSS settings:
-application detection level: Low
-framerate limit: 138
-on screen display rendering mode: vector 3D
-on screen display coordinate space: viewport
-game applying the vsync vs NVCP applying the vsync
-All PCI devices in message signaled interrupt mode (
http://forums.guru3d.com/showthread.php?t=378044)
I suspect the application detection level is affecting input lag levels, and I believe 142 is not a sufficiently low to keep it in the gsync range 100.0% of the time either. The "raster 3d" mode of on screen display is more advanced and supports more fonts which could potentially take longer to render as well.
Having said this, I find myself surprised to seemingly be detecting differences in input lag with RTSS vs in-game limiter. It only seems to occur on asset loading, such as when getting to a new section of a map, where I think I can feel the input lag growing at the mouse cursor for a few seconds and then settling back to normal after a few seconds, whereas with the in-game limiter I seem to not be able to detect this. It's so hard to tell objectively though, and you are one of only two guys with the objective tools. It's a problem in iracing for example because even the slightest deviation of input lag , even briefly of 7ms, can cause you to miss your turn in point by a few meters which amounts to actually a big loss in time. It would be better if it actually stayed at 7ms then you could adjust to it, but when it's spiking randomly +/-7m you can't compensate for that, which RTSS seems to be exacerbating. I read somewhere that it's because RTSS is stalling the whole rendering pipeline whereas the in-game limiter is only sleeping inside the render thread while other threads are still free to do things like calculate the next scene etc.
The delay RTSS introduces has nothing to do with the RTSS, in-game or control panel settings. RTSS simply doesn't begin regulation of the framerate early enough in the rendering cycle, and it couldn't unless it was part of the game engine it was being used in.
I'm currently very busy finishing up my upcoming G-SYNC 101: Input Latency article, so I can assure you, that with G-SYNC, 2 frames below the refresh rate is already safe, and 3 frames is super safe.
I will be including an in-game vs RTSS vs Nvidia Inspector FPS limiter test, but only for one game. Perhaps I can dedicate a future article to testing RTSS across a series of games, but that would come much later.
As for the "PCI devices in message signaled interrupt mode," that's really getting into the weeds, and as far as I know, with Windows 10, IRQ settings and conflicts are a thing of the past, and it is all automatically managed by the system.
pneu wrote:jorimt wrote:
@RealNC, The Witcher 3 has an asset streaming system that is prone to creating frequent frametime spikes (at least on initial load)
I hope you are running it off SSD because I have noticed the same thing on Witcher 2 SSD vs HDD, where the former is free of frame spikes on asset loading, or, the latter is free of spikes but only after the first run through the map, because on the second run through windows is reading it from its memory cache, so it actually loads from RAM the second time (or possibly the cache on the HDD itself - but not superfetch/prefetch because I have those disabled). Observe similar thing in other games - first run through everything is loading off disk and disk reads have latency of ~25ms for a HDD vs 4ms for a SSD (according to Windows Resource Monitor) but second run through it's all from memory and much faster, no pauses when loading new sections of the map.
I've run it off an SSD and an HDD, and no change in frametime spikes due to asset load, but loading screens are considerably shortened, yes, which is the typical improvement seen in an SSD.
The Witcher 3 appears to stream in most of its assets upon initial load-up over a period of a few minutes, which then appear to cache, unless the game is restarted, or a save is manually reloaded. So while an SSD may load the initial assets in faster, it still has to load them in, which results in hitches here and there regardless, at least in my experience.