Blur Busters Forums

Here is the article:
https://www.nvidia.com/en-us/geforce/ne ... dy-driver/

It looks NVIDIAs new driver offers lots of exciting functionality. Although their "Ultra low latency" mode and new sharpening filter seem to be ideas directly stolen from AMD, it's obviously still the best for consumers, if we all have access to them. I'm sure we're gonna see some comparisons soon from HardwareUnboxed and others. The best part is that the features are available on all cards, DX9 + DX11 for latency and DX12 on top for the sharpening.
Edit: I've installed the driver and the low latency feature doesn't seem to be there. It still shows "Maximum pre-rendered frames" as before with the same options.

Integer scaling is another nice feature that has been requested for a while. It's only available for Turing cards. I'd assume that's because they can do those integer + floating point operations at the same time and can do the scaling with no performance loss most of the time? Just a wild guess, but I'm not an expert. Might also be only to push the newer GPUs.

There are some further improvements in the driver so make sure to check out the article for the full information.

I got the low latency thing on my GPU, double check if you have it.
https://i.imgur.com/P2TK35O.png (proof)
havent tested anythin

Vechs wrote:I got the low latency thing on my GPU, double check if you have it.
https://i.imgur.com/P2TK35O.png (proof)
havent tested anythin

Thanks. I just tried a clean reinstall and now it is there. Edit: I should note that the driver window was claiming to already be on the correct version before the reinstall. Must've just had some problem during the installation, I guess.

Finally, about time. I knew that Max Prerendered 0 needed to come back.

If they implemented it properly, it may have equivalent latency as RTSS scanline sync.

Rob (aka Gridl0ck) has now posted this on the main Blur Busters news stream now.

New NVIDIA Graphics Driver Cuts Input Latency

It sounds the same as scanline sync to me. Is there a difference?

lossofmercy wrote:It sounds the same as scanline sync to me. Is there a difference?

The technique is definitely different but the latency might be same or similar.

Depending on how it is done, algorithmically.

RTSS Scanline Sync bypassed the max prerendered frames by using VSYNC OFF to emulate VSYNC ON via beam-racing the tearlines offscreen, and avoiding prerendered frames. RTSS Scanline Sync was a rube-goldberg defacto equivalent of Max Prerendered Frames 0, dones as a software based bypass/workaround.

Proper driver-based implementation would be, hopefully, vastly superior. Latency needs to be measured.

Forgive me if I am not understanding this, but I am a little lost.

I am translating pre-rendered frames to "buffered frame." IE, double buffer, triple buffer in the VSYNC on.

I thought the whole point of buffering was that you can have a frame "in stock" if the new frame isn't done by the time the monitor updates with V-SYNC on. What you are doing with smart-sync is making the timing of checking the buffer outside of the "frame" or the monitor itself, making tearing invisible to the user, right?

What it sounds like Nvidia is doing with max pre-rendered frame is that they are removing the buffer frame... but I don't understand how it's any different than a regular v-sync? And would that create issues with stuttering if it ever missed a frame?

Tried integer scaling on my RTX 2060... seems like it doesn't work yet.
Behaves as "no scaling".

lossofmercy wrote:Forgive me if I am not understanding this, but I am a little lost.

I am translating pre-rendered frames to "buffered frame." IE, double buffer, triple buffer in the VSYNC on.

Prerendered frames are something else. What you're referring to are rendered frames. You have two rendered frames when doing double buffering, three when doing triple buffering. Prerendered frames are frame data that has not been rendered yet. Rendered frames are created by the GPU. Prerendered frames are created by the CPU and then passed to the GPU. The GPU takes prerendered frames from the CPU and turns them into rendered frames.

This system makes better use of the CPU. While the GPU is rendering the current frame, the CPU will prepare the data for the next frame (prepare a prerendered frame). It will read input from the player, calculate the new state of the world, do the physics calculations, and everything else that is needed to start rendering a new frame. If the GPU is still busy rendering the current frame, the CPU will then proceed to prerender yet another frame. So the CPU runs ahead of the GPU if the GPU can't render fast enough. How much the CPU runs ahead of the GPU depends on the value of "max prerendered frames."

This new "ultra" setting will prevent the CPU from prerendering even 1 such data set until the GPU has mostly finished rendering the current frame so that player input is read as late as possible (which is why input lag is reduced.) This is also why it can affect performance negatively, because the CPU now has less time to prerender the next frame. So this setting works best in GPU-limited games, because then the CPU is not the bottleneck and thus starts to run ahead of the GPU (which is the bottleneck) most of the time.

Very nice explanation.