that's why my pc is in cold storage and i bought a ps5
ready to make a return though, if the sillyness ends.
-
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.
RDNA 3 rumours Q3/4 2022
- Thread starter Grim5
- Start date
- Status
ready to make a return though, if the sillyness ends.
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,219
So far nothing for DLSS 3.x which is to be expected. Hopefully FSR will evolve fast and good enough.
We don't know the actual effect upon input latency and since not everyone plays "competitively", at hundreds of FPS, it probably doesn't matter that much. After all, some people are/were fine to play on a gamepad and at even at 30fps, so talking about input lag above all doesn't hold much weight.
Caporegime
If DLSS 3 takes data from the next frame to fill in the detail on the low res frame it puts on your screen how does it do that? Wouldn't it need to render the next frame to get that detail?
Caporegime
So you are saying ...
Jensen is disappointed in you.
I tried to think of something witty, I have nothing.
Soldato
Which in itself is quite witty.
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,219
I guess it can render then next frame, but not display it. Then you add the frame(s) in between and start displaying the frames.
There are commands in certain games/engine like Render Ahead_x (I think it was, probably still is in nVIDIA drivers), where x being a higher number meant a smoother experience at the cost of some input lag, while lower values meant lower input lag, bug increased stutter. Perhaps works in a similar way, but now the hardware has enough power to generate frames between those reference ones to increase smoothness ( there was a talk that for each "real" pixe 7 more are generated by AI?).
In a way, like that works v-sync if I'm not mistaken, as in holding back/generating just enough frames to match the display refresh rate resulting in a slight lag - sometimes can be bigger, depends per game.
Last edited:
Soldato
Sad state of affairs really even more so when you think most games now are rubbish - maybe 1-2 a year that are really good if lucky.
Maybe just full of snobs on this forum and have much better kit than the average but hard not to see the death of pc gaming if this continues, crazy but now seems sensible £700ish for 1080Ti now more than double for high end.
Maybe just full of snobs on this forum and have much better kit than the average but hard not to see the death of pc gaming if this continues, crazy but now seems sensible £700ish for 1080Ti now more than double for high end.
Caporegime
The point is it would be using GPU cycles to render a frame its not displaying, displaying the frame costs nothing, rendering it costs GPU cycles, the same as if it was putting that image on your screen. Its rendering a higher res image to gather data from, not to display on the screen.
So does Jensons explanation whats going on here make any sense to you? It doesn't to me.
Last edited:
I do one better, people paying this money to wait on games released on the ps5 5 years later, while many can get those games now, looks great, performs great and on discount lol
Soldato
Yep true Forbidden west was good on PS5 i did miss the mouse for aiming though but single player so once completed a few times thats it
Soldato
Wont need a new PSU either..
From memory I believe the popular rumour is that the main die is about 300 mm2
Love the size of my MBA 6950XT - hoping the 7900 is a similar form factor, for the reference card at least. Judging from the picture they showed before it looks like an evolution of the current design so hopefully the same size!
My 6950 fits perfectly in a circa 20L SFF case (Tecware fusion). Where as something like a 4090 just won’t work - unless water cooled.
My 6950 fits perfectly in a circa 20L SFF case (Tecware fusion). Where as something like a 4090 just won’t work - unless water cooled.
Last edited:
Soldato
The optimist in me FTFY both.
Last edited:
Caporegime
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,219
From memory: AI is trained on supercomputers using high res images, that was/is (?) the model for current versions. Using that it can accurately "guess" how to fill image properly when it upscale it.
So basically everything is the same until this point. You have frame 1 and 2, but you have them in vRAM, you don't display them. Using their new tech, they can quickly calculate and add the other frame needed to fit between the original 1 and 2 frame. For every "real" pixel of the frame, 7 other pixels are "guessed" by the GPU and this "guessing" is far less of a performance hog than it would be to render the entire frame.
So, since the magic from DLSS is already working and we can see it in action in many games, the other part of magic is creating the other frame that is slided between other two. Obviously I don't know how well that works, but I'm guessing pretty good since DLSS only got better with each iteration.
In terms of latency, I'd say is more about the overall latency: mouse, computer (with all its hardware and software bits), plus display. Each adds naturally its own lag.
Lets say you have 120-130ms for a given scenario from the moment you press the mouse button until the frame is fully rendered on screen. From that, 50ms is down to the GPU and drivers. Since that is not a hard wall and subject to improvement (reflex already does that and AMD also I believe does this), you cut that 50ms to 40ms. If the time needed to generate the extra frame is 10ms, then the latency is the same. If is less, then you have a gain. If is more, well, if is 20ms then you go from the original 50ms to 60ms. If is 30ms then is 70ms and so on. But, the new overall latency would be 130-140ms, etc., so potentially irrelevant.
Granted, until this is fully tested (for overall latency: mouse-pc-display), we can only guess. But if someone half tests this (just the GPU and maybe driver latency), then it can give a false negative.
I'm hoping AMD can do something similar rather sooner than later with FSR. Having something open and available to all is a indeed a much better option.
Man British shoebox houses are so small lmao
- Status