Permabanned
Essentially your GPU is lying to you.
100 FPS at 50ms latency is not at all a good experience.
100 FPS at 50ms latency is not at all a good experience.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Will DLSS 3 add latency ?
- Thread starter Donnie Fisher
- Start date
I noticed that Nvidia actually put the times on a couple of its charts and one had DLSS2 and one has DLSS3. So this is our first look at how much latency DLSS3 adds.
Short answer: latency is still lower than just using DLSS2 with a lower framerate.
Here is DLSS2 in cyberpunk, game is doing 62fps, latency is 58ms
Here is DLSS3 turned on, framerate jumps from 62fps to 101fps and latency drops from 58ms to 55ms
Based on this, I'm going to go out on a limb and suggest that in games where your framerate is already very high, you probably don't want to use DLSS3 - so like you wouldn't want to use it in Overwatch 2 when you're already on 400fps. But in graphically demanding ray tracing games, it looks like you can boost framerate higher than DLSS2 and still get lower latency with that higher framerate - the latency is not as low as if you were getting 101fps natively but you're never going to get that in demanding games anyway.
Short answer: latency is still lower than just using DLSS2 with a lower framerate.
Here is DLSS2 in cyberpunk, game is doing 62fps, latency is 58ms
Here is DLSS3 turned on, framerate jumps from 62fps to 101fps and latency drops from 58ms to 55ms
Based on this, I'm going to go out on a limb and suggest that in games where your framerate is already very high, you probably don't want to use DLSS3 - so like you wouldn't want to use it in Overwatch 2 when you're already on 400fps. But in graphically demanding ray tracing games, it looks like you can boost framerate higher than DLSS2 and still get lower latency with that higher framerate - the latency is not as low as if you were getting 101fps natively but you're never going to get that in demanding games anyway.
Last edited:
Seems odd to me - there definitely wasn't anything like 50ms latency in CP2077 with DLSS on my setup.
I had one of the early 4K monitors which had around 50-60ms latency (possibly even a little higher than that) between the start of the frame rendering and the pixel fully changing on the display and the experience for gaming was terrible.
Last edited:
It's probably measuring total system latency from mouse click to pixel completion, not just the screen's latency. This is the way Nvidia has been measuring latency in all its other latency marketing material in the last while
Last edited:
Even then it seems rather high, though inline with their video, my setup is around 9ms for system + whatever screen latency - usually sub 16ms total when using a high performance gaming monitor.
Last edited:
so if you move your mouse 1 pixel, then 1 pixel more and stop Nvidia will move your mouse 1 pixel again before realising it is in the wrong place.
you'll need your sea legs for DLSS3.0
Soldato
What does the nvidia perf overlay show render latency at for you ?
At around 62 fps I see somewhere between 22 and 30ms in CP2077
Don't have CP2077 installed on my main system currently.
If that is what they are using for measuring DLSS 3, then it is flawed method of measuring latency in this instance.
When gaming it is a closed loop. You see something on screen, click, it is processed and something new on screen pops up.
The previous method has been fine but with Nvidia now changing the rate that game information is displayed on screen it is no longer a good method. Hopefully reviewers will pick up on this.
Man of Honour
People will need to understand that they will need a mouse that is also Reflex compatible for all this to work hand in hand. All this is adding up to an already expensive affair. New crazy GPU price, new PSU needed most likely, new gaming mouse needed.
£££.
Reflex just reduces frame render ahead doesn't it? You should be able to do the same thing to games without DLSS if latency is important (and you can afford to not render ahead) - no mouse requirements at all.
I hadn't pick up on DLSS frame interpolation working on previous frames rather than rendered ahead frames - the latter would obviously be incompatible with Reflex, but if it's previous frames then render ahead would actually end up adding even more frames between inputs so makes sense that you actually want Reflex on when using DLSS frame interpolation - it might even become a requirement.
My take on the OP: no, it won't add latency, but you also don't gain any improvement in latency from increased framerates (many people like improved framerates precisely because of the effect of reduced latency, but that won't be the case with these interpolated frames).
I hadn't pick up on DLSS frame interpolation working on previous frames rather than rendered ahead frames - the latter would obviously be incompatible with Reflex, but if it's previous frames then render ahead would actually end up adding even more frames between inputs so makes sense that you actually want Reflex on when using DLSS frame interpolation - it might even become a requirement.
My take on the OP: no, it won't add latency, but you also don't gain any improvement in latency from increased framerates (many people like improved framerates precisely because of the effect of reduced latency, but that won't be the case with these interpolated frames).
Last edited:
Why would anynone use DLSS on an esport? They usually select the lowest graphics settings and crack on.
Caporegime
This doesn't follow.
Without frame generation, frame rate is 50Hz, a frame every 20ms and input polled every 20ms.
With frame generation, frame rate is 100Hz, a frame every 10ms and input polled every 20ms.
The latency remains the same.
Have a look at Nvidia's diagram:
The generated frame comes after the rendered one, not before it.
I think this probably depends on how it is integrated with the game engine. If the engine has full control and access to what is happening with these generated frames then it can use the headroom allowed by half the frames being generated to run a physics update on every rendered frame, doubling the number of times it polls input each second. Obviously you won't see this on screen immediately, but you would get the faster response.
Last edited:
Are you sure game engines only poll player inputs at the same rate as the frame rate?
Just seen the latter half of your post. I'm sure i heard something about games decoupling physics from frame rate other wise the game speeds up as framerate increases.
Last edited:
Caporegime
Yeah, you're right, a lot of games do work that way, so frame rate won't affect latency at all.
Associate
the only way you'd get low latency is through just in time rendering where the game doesnt maintain a queued buffer of frames.. which also translates to low fps, i dont think any kind of increased activity in the game world can negate that
also the absolute order of frame rendering shouldnt matter, whether it is the first or second frame, because when you look at it in a sequence theres always a frame buffer being queued which would impact latency
whats interesting is how quickly the new method can reorder/flush the buffer queue to deliver a better experience, sounds interesting
also the absolute order of frame rendering shouldnt matter, whether it is the first or second frame, because when you look at it in a sequence theres always a frame buffer being queued which would impact latency
whats interesting is how quickly the new method can reorder/flush the buffer queue to deliver a better experience, sounds interesting
It's not really about polling, but when the data is last updated to make use of said polling. Assuming you remove frames rendered ahead (ie Reflex) then you update the simulation each time just before you output the information - higher frame rates means more updates per second therefore it feels like it's polling more often. When you add in frames rendered ahead, or DLSS frame interpolation, you aren't changing the rate that you update the simulation, despite (especially in the latter case) outputting more frames - though if it predicted accurately enough it might feel like it did!
Last edited:
Nvidia adding fake frames in because they know they won't be able to keep up with the raw power of the 7900XT 
couldn't make it up. They might as well just make there own fps counting software that reports double the frame rate
couldn't make it up. They might as well just make there own fps counting software that reports double the frame rateThey could add some transparent frames to fool the fps countersNvidia adding fake frames in because they know they won't be able to keep up with the raw power of the 7900XT
And in quite a few cases, dlss will have better motion clarity than native/TAA but obviously if you're setting everything to low4pro then native/TAA won't be an issue.