Some links:
Many G-SYNC input lag tests & graphs -- on a 240Hz eSports monitor -- via 2 months of testing via high speed video!
www.blurbusters.com
Contribute to Orbmu2k/nvidiaProfileInspector development by creating an account on GitHub.
github.com
Speaking of gsync, is leaving vsyinc on in game correct ?
When using G-SYNC, vsync generally does nothing as long as the frame rate is within the displays VRR range. If your frame rate exceeds the maximum refresh rate of the display, you can encounter tearing as the frame rate is not capped unless vsync is enabled, or you have a frame rate limiter. You may also encounter tearing near the top end of the VRR range as synching will rapidly engage and disengage with fluctuating frame times. Vsync at max refresh rate can also introduce considerable (relative to frame times) latency. This is why many people enable vsync and set a frame rate cap slightly below (2-3 frames) the max refresh rate of the display to ensure there is no tearing and that the VRR range is never exceeded.
Personally, since I do not want to cap my frame rate (many things in ED are frame rate sensitive), but I also wish to completely avoid tearing (it's the whole reason I've got a VRR display), I leave vsync off in-game and enable "fast" vsync in the NVIDIA control panel, along with G-SYNC.
Is there some particular setting i should change from default in the nvidia control panel ?
I use high quality texture filtering, force 16x AF, enable fast sync, and set low latency mode to ultra. The power management mode should pretty much always be set to "Prefer maximum performance" on a desktop.
I also force threaded optimization on. I'm pretty sure on is the default for ED, but you'd only want to disable it if you have a limited number of very fast cores (e.g. a 5GHz quad core might benefit from it, but something with more cores will not).
Additionally, I use NVIDIA Profile Inspector (the one by
Orbmu2k) to disable the CUDA P2 state (by default NVIDIA's drivers kick the memory clock down when running CUDA code, which includes NVENC, their hardware video encoder, significantly increasing it's performance hit) and force the default "memory allocation policy" to "moderate preallocation", which seems to smooth out ED ever so slightly.
What about low latency mode ?
This is another name for render queue depth, flip queue size, or maximum frames render ahead.
Most D3D games default to a queue depth of three, meaning it can prepare this many frame a head of time to smooth out rendering pauses...good for smoothness and frame rate, bad for latency.
If you aren't CPU limited (and you should not be), I recommend setting this to "Ultra" which sets the queue depth to zero, not prerendering any frames and resulting in the lowest latency possible.
As of driver version 441.08 these low latency settings work correctly with G-SYNC.
