Jump to content

Double your FPS for FREE!*** - Nvidia DLSS 3.0

BellLMG
By BellLMG
in LTT Releases
 Share
Posted

 

Nvidia’s new DLSS 3.0 promises smoother gameplay by doubling your FPS – But how can it do that, and can its fake frame generation possibly look any good? And is it worth buying an RTX 4090 for?

 

 

Link to comment
Share on other sites
Link to post
Share on other sites
Posted

Leave it to LinusMediaGroup to release a 17:60 minute video without understanding even how the underlying technology works.

Please watch the GTC keynote, where Nvidia founder Jensen Huang clearly explains it to you once again.

Currently this video is just further spreading misinformation, causing people confusion.

 

 

Your explanation from 2:09 onwards is completely WRONG. Like complete misinformation.

DLSS 3.0 does not make an educated guess what the "next frame" should look like based on the previous frame and motion vectors.

IT makes an educated guess what the intermediate frame should be based on the previous frame, motion vectors AND THE CURRENT FRAME.

Here is an transcript fron J. Huang himself:

Spoiler
10:00
Ada introduces DLSS 3, a new AI that generates entirely new frames, rather than just pixels.
10:10
DLSS 3 has four components - a new Optical Flow Accelerator, game engine motion vectors, a convolutional autoencoder AI
10:20
frame generator, and our Reflex super-low-latency pipeline.
10:25
DLSS 3 processes the new frame and the prior frame to discover how the scene is changing.
10:32
The Optical Flow Accelerator provides the neural network with the direction and velocity of pixels from frame to frame.
10:40
Pairs of frames from the game, along with the geometry and pixel motion vectors, are then fed into a neural network

IT GENERATES THE INTERMEDIATE FRAMES. Not a future frame. 

IF the GPU has generated 2 real frames, it generates a fake frame between the 2 real frames. This is where the inherit question about latency comes in. The next frame has to be rendered before it can show the generated frame between 2 frames, so it holds back the "current" frame, until the DLSS 3.0 frame is rendered.

If you are still have trouble understanding the pipeline, here is the dumbed down version for you:

Spoiler

1) gpu renders 1 frame, then displays it.

2) gpu renders 2nd frame, doesn't display it, instead feeds it into the neural net.

3)gpu generates a fake frame, displays it.

4)gpu finally displays the 2nd frame.

etc

 

Here is what you claimed in your video:

Spoiler

Gpu renders 1 frame-> displays it.

Based on frame 1 and motion vectors, gpu extrapolates a fake frame and displays it.

gpu renders 2nd frame, displays it.

etc

 

You might be saying that I am nitpicking, but the conceptional difference between the way it works, and the way it actually works is MASSIVE, and has way different implications about input latency as well as how close the generated frame could be to "perfect frame" as well.

@BellLMG, @GabenJr Please pull the video and correct it, or at the very least pin a top comment explaining how the technology actually works. Because what you have currently done is just causing more confusion about the technology, not clarity.

You already have caused confusion on reddit: 

https://www.reddit.com/r/virtualreality/comments/yphgzz/dlss3_frame_generation_actually_extrapolation/

 

I only see your reply if you @ me.

This reply/comment was generated by AI.

Link to comment
Share on other sites
Link to post
Share on other sites
Posted
8 minutes ago, Origami Cactus said:

 @GabenJr Please pull the video and correct it, or at the very least pin a top comment explaining how the technology actually works. Because what you have currently done is just causing more confusion about the technology, not clarity.

You already have caused confusion on reddit: 

https://www.reddit.com/r/virtualreality/comments/yphgzz/dlss3_frame_generation_actually_extrapolation/

 

Just to note they are probably asleep, it's like 5:50 am in Vancouver.

Link to comment
Share on other sites
Link to post
Share on other sites
Posted
5 minutes ago, Ultraforce said:

Just to note they are probably asleep, it's like 5:50 am in Vancouver.

Yeah, I know about the timezone difference, but that doesn't make spreading misinformation any more acceptable. I just wrote the reply now, and hopefully they see the reply because I quoted them, and then act accordingly.

But how such an underlying mistake about the technology got past all the millions of people that LMG employs is past me. Many different articles and videos about the subject have gotten it correct, so they had mountains of evidence that their conclusion is completely false.

I have to say it again. They missed the fact that the neural net takes previous and future frame as an input to render a frame between them, it doesn't only take a previos frame to extrapolate a new "future" frame.

If someone wants to watch a video that actually had at least some reasearch put into it, here is the Hardware unboxed video from 3 weeks back about the same subject:

https://www.youtube.com/watch?v=GkUAGMYg5Lw

 

I only see your reply if you @ me.

This reply/comment was generated by AI.

Link to comment
Share on other sites
Link to post
Share on other sites
Posted

Yeah I don't know about this video. Other users have highlighted great points and the title is incredibly misleading if not just a lie. Putting asterisks on a word doesn't just mean you can mislead people. The larger LTT gets the worse they look using clickbait and having misleading videos

▓ƆԀ▓ S▓₱▓Ɇ▓c▓s▓: i7 6ʇɥפᴉƎ00K (4.4ghz), Asus DeLuxe X99A II, GT҉X҉1҉0҉8҉0 Zotac Amp ExTrꍟꎭe),Si6F4Gb D???????r PlatinUm, EVGA G2 Sǝʌǝᘉ5ᙣᙍᖇᓎᙎᗅᖶt, Phanteks Enthoo Primo, 3TB WD Black, 500gb 850 Evo, H100iGeeTeeX, Windows 10, K70 R̸̢̡̭͍͕̱̭̟̩̀̀̃́̃͒̈́̈́͑̑́̆͘͜ͅG̶̦̬͊́B̸͈̝̖͗̈́, G502, HyperX Cloud 2s, Asus MX34. פN∩SW∀S 960 EVO

Just keeping this here as a backup 9̵̨̢̨̧̧̡̧̡̧̡̧̡̡̢̢̡̢̧̡̢̡̡̢̧̛̛̛̛̛̛̱̖͈̠̝̯̹͉̝̞̩̠̹̺̰̺̲̳͈̞̻̜̫̹̱̗̣͙̻̘͎̲̝͙͍͔̯̲̟̞͚̖̘͉̭̰̣͎͕̼̼̜̼͕͎̣͇͓͓͎̼̺̯͈̤̝͖̩̭͍̣̱̞̬̺̯̼̤̲͎̖̠̟͍̘̭͔̟̗̙̗̗̤̦͍̫̬͔̦̳̗̳͔̞̼̝͍̝͈̻͇̭̠͈̳͍̫̮̥̭͍͔͈̠̹̼̬̰͈̤͚̖̯͍͉͖̥̹̺͕̲̥̤̺̹̹̪̺̺̭͕͓̟̳̹͍̖͎̣̫͓͍͈͕̳̹̙̰͉͙̝̜̠̥̝̲̮̬͕̰̹̳͕̰̲̣̯̫̮͙̹̮͙̮̝̣͇̺̺͇̺̺͈̳̜̣̙̻̣̜̻̦͚̹̩͓͚̖͍̥̟͍͎̦͙̫̜͔̭̥͈̬̝̺̩͙͙͉̻̰̬̗̣͖̦͎̥̜̬̹͓͈͙̤̜̗͔̩̖̳̫̑̀̂̽̈́̈́̿͒̿̋̊͌̾̄̄̒̌͐̽̿̊͑̑̆͗̈̎̄͒̑̋͛̑͑̂͑̀͐̀͑̓͊̇͆̿͑͛͛͆́͆̓̿̇̀̓͑͆͂̓̾̏͊̀̇̍̃́̒̎̀̒̄̓̒̐̑̊̏̌̽̓͂͋̓̐̓͊̌͋̀̐̇̌̓̔͊̈̇́̏͒̋͊̓̆̋̈̀̌̔͆͑̈̐̈̍̀̉̋̈́͊̽͂̿͌͊̆̾̉͐̿̓̄̾͑̈́͗͗̂̂́̇͂̀̈́́̽̈́̓̓͂̽̓̀̄͌̐̔̄̄͒͌̈́̅̉͊̂͒̀̈́̌͂̽̀̑̏̽̀͑̐̐͋̀̀͋̓̅͋͗̍́͗̈́̆̏̇͊̌̏̔̑̐̈́͑̎͑͆̏̎́̑̍̏̒̌̊͘͘̚̕̚̕̕̚̕̚̕̕͜͜͜͜͜͝͝͠͠͝͝͝͝͝͝͝͠͝͝ͅͅͅͅͅͅͅ8̵̨̛̛̛̛̮͍͕̥͉̦̥̱̞̜̫̘̤̖̬͍͇͓̜̻̪̤̣̣̹̑͑̏̈́̐̐́̎͒̔͒̌̑̓̆̓͑̉̈́́͋̌͋͐͛͋̃̍̽̊͗͋͊̂̅͊͑́͋͛̉̏̓͌̾̈́̀͛͊̾͑̌̀̀̌̓̏̑́̄̉̌͂́͛̋͊̄͐͊̈́̀̌̆̎̿̓̔̍̎̀̍̚̕̕͘͘͘̕̚͝͝͠͠͠0̶̡̡̡̢̨̨͕̠̠͉̺̻̯̱̘͇̥͎͖̯͕̖̬̭͔̪̪͎̺̠̤̬̬̤̣̭̣͍̥̱̘̳̣̤͚̭̥͚̦͙̱̦͕̼͖͙͕͇̭͓͉͎̹̣̣͕̜͍͖̳̭͕̼̳̖̩͍͔̱̙̠̝̺̰̦̱̿̄̀͐͜͜ͅͅt̶̡̨̡̨̧̢̧̢̨̧̧̧̧̢̡̨̨̢̨̢̧̢̛̛̛̛̛̠͍̞̮͇̪͉̩̗̗͖̫͉͎͓̮̣̘̫͔̘̬̮̙̯̣͕͓̲̣͓͓̣̹̟͈̱͚̘̼̙̖̖̼̙̜̝͙̣̠̪̲̞̖̠̯̖̠̜̱͉̲̺͙̤̻̦̜͎̙̳̺̭̪̱͓̦̹̺͙̫̖̖̰̣͈͍̜̺̘͕̬̥͇̗̖̺̣̲̫̟̣̜̭̟̱̳̳̖͖͇̹̯̜̹͙̻̥̙͉͕̜͎͕̦͕̱͖͉̜̹̱̦͔͎̲̦͔̖̘̫̻̹̮̗̮̜̰͇̰͔̱͙̞̠͍͉͕̳͍̰̠̗̠̯̜̩͓̭̺̦̲̲͖̯̩̲̣̠͉̦̬͓̠̜̲͍̘͇̳̳͔̼̣͚̙͙͚͕̙̘̣̠͍̟̪̝̲͇͚̦̖͕̰̟̪͖̳̲͉͙̰̭̼̩̟̝̣̝̬̳͎̙̱͒̃̈͊̔͒͗̐̄̌͐͆̍͂̃̈́̾͗̅̐͒̓̆͛̂̾͋̍͂̂̄̇̿̈͌̅̈́̃̾̔̇̇̾̀͊͋̋̌̄͌͆͆̎̓̈́̾̊͊̇̌̔̈́̈́̀̐͊̊̍͑̊̈̓͑̀́̅̀̑̈́̽̃̽͛̇́̐̓̀͆̔̈̀̍̏̆̓̆͒̋́̋̍́̂̉͛̓̓̂̋̎́̒̏̈͋̃̽͆̓̀̔͑̈́̓͌͑̅̽́̐̍̉̑̓̈́͌̋̈́͂̊́͆͂̇̈́̔̃͌̅̈́͌͛̑̐̓̔̈́̀͊͛̐̾͐̔̾̈̃̈̄͑̓̋̇̉̉̚̕̚͘̕̚̚̕̕͜͜͜͜͜͜͜͜͜͜͜͜͜͝͝͝͠͝͝͝͝͝͠ͅͅͅͅͅi̵̢̧̢̧̡̧̢̢̧̢̢̢̡̡̡̧̧̡̡̧̛̛͈̺̲̫͕̞͓̥̖̭̜̫͉̻̗̭̖͔̮̠͇̩̹̱͈̗̭͈̤̠̮͙͇̲͙̰̳̹̲͙̜̟͚͎͓̦̫͚̻̟̰̣̲̺̦̫͓̖̯̝̬͉̯͓͈̫̭̜̱̞̹̪͔̤̜͙͓̗̗̻̟͎͇̺̘̯̲̝̫͚̰̹̫̗̳̣͙̮̱̲͕̺̠͉̫̖̟͖̦͉̟͈̭̣̹̱̖̗̺̘̦̠̯̲͔̘̱̣͙̩̻̰̠͓͙̰̺̠̖̟̗̖͉̞̣̥̝̤̫̫̜͕̻͉̺͚̣̝̥͇̭͎̖̦̙̲͈̲̠̹̼͎͕̩͓̖̥̘̱̜͙̹̝͔̭̣̮̗̞̩̣̬̯̜̻̯̩̮̩̹̻̯̬̖͂̈͂̒̇͗͑̐̌̎̑̽̑̈̈́͑̽́̊͋̿͊͋̅̐̈́͑̇̿̈́̌͌̊̅͂̎͆̏̓͂̈̿̏̃͑̏̓͆̔̋̎̕͘͘͘͜͜͜͜͜͜͜͝͝͠͠ͅͅͅͅͅͅͅͅͅZ̴̧̢̨̢̧̢̢̡̧̢̢̢̨̨̨̡̨̧̢̧̛̛̬̖͈̮̝̭̖͖̗̹̣̼̼̘̘̫̠̭̞͙͔͙̜̠̗̪̠̼̫̻͓̳̟̲̳̻̙̼͇̺͎̘̹̼͔̺̹̬̯̤̮̟͈̭̻͚̣̲͔͙̥͕̣̻̰͈̼̱̺̤̤͉̙̦̩̗͎̞͓̭̞̗͉̳̭̭̺̹̹̮͕̘̪̞̱̥͈̹̳͇̟̹̱̙͚̯̮̳̤͍̪̞̦̳̦͍̲̥̳͇̪̬̰̠͙͕̖̝̫̩̯̱̘͓͎̪͈̤̜͎̱̹̹̱̲̻͎̖̳͚̭̪̦̗̬͍̯̘̣̩̬͖̝̹̣̗̭͖̜͕̼̼̲̭͕͔̩͓̞̝͓͍̗̙̯͔̯̞̝̳̜̜͉̖̩͇̩̘̪̥̱͓̭͎͖̱̙̩̜͎̙͉̟͎͔̝̥͕͍͓̹̮̦̫͚̠̯͓̱͖͔͓̤͉̠͙̋͐̀͌̈́͆̾͆̑̔͂͒̀̊̀͋͑̂͊̅͐̿́̈́̐̀̏̋̃̄͆͒̈́̿̎́́̈̀̀͌̔͋͊̊̉̿͗͊͑̔͐̇͆͛̂̐͊̉̄̈́̄̐͂͂͒͑͗̓͑̓̾̑͋̒͐͑̾͂̎̋̃̽̂̅̇̿̍̈́́̄̍͂͑̏̐̾̎̆̉̾͂̽̈̆̔́͋͗̓̑̕͘̕͘͜͜͜͜͜͝͝͝͝͠͠͝ͅo̶̪͆́̀͂̂́̄̅͂̿͛̈́̿͊͗́͘͝t̴̡̨̧̨̧̡̧̨̡̢̧̢̡̨̛̪͈̣̭̺̱̪̹̺̣̬̖̣̻͈̞̙͇̩̻̫͈̝̭̟͎̻̟̻̝̱͔̝̼͍̞̼̣̘̤̯͓͉̖̠̤͔̜̙͚͓̻͓̬͓̻̜̯̱̖̳̱̗̠̝̥̩͓̗̪̙͓̖̠͎̗͎̱̮̯̮͙̩̫̹̹̖͙̙͖̻͈̙̻͇͔̙̣̱͔̜̣̭̱͈͕̠̹͙̹͇̻̼͎͍̥̘͙̘̤̜͎̟͖̹̦̺̤͍̣̼̻̱̲͎̗̹͉͙̪̞̻̹͚̰̻͈͈͊̈́̽̀̎̃̊́̈́̏̃̍̉̇̑̂̇̏̀͊̑̓͛̽͋̈́͆́̊͊̍͌̈́̓͊̌̿̂̾̐͑̓̀́͒̃̋̓͆̇̀͊̆͗̂͑͐̀͗̅̆͘̕͘̕̕͜͜͝͝͝͝͝͝͝ͅͅͅͅͅͅͅͅͅḁ̶̢̡̨̧̡̡̨̨̧̨̡̡̢̧̨̡̡̛̛̛͍̱̳͚͕̩͍̺̪̻̫̙͈̬͙̖͙̬͍̬̟̣̝̲̼̜̼̺͎̥̮̝͙̪̘̙̻͖͇͚͙̣̬̖̲̲̥̯̦̗̰̙̗̪̞̗̩̻̪̤̣̜̳̩̦̻͓̞̙͍͙̫̩̹̥͚̻̦̗̰̲̙̫̬̱̺̞̟̻͓̞͚̦̘̝̤͎̤̜̜̥̗̱͈̣̻̰̮̼̙͚͚̠͚̲̤͔̰̭̙̳͍̭͎̙͚͍̟̺͎̝͓̹̰̟͈͈̖̺͙̩̯͔̙̭̟̞̟̼̮̦̜̳͕̞̼͈̜͍̮͕̜͚̝̦̞̥̜̥̗̠̦͇͖̳͈̜̮̣͚̲̟͙̎̈́́͊̔̑̽̅͐͐͆̀͐́̓̅̈͑͑̍̿̏́͆͌̋̌̃̒̽̀̋̀̃̏̌́͂̿̃̎̐͊̒̀̊̅͒̎͆̿̈́̑̐̒̀̈́̓̾͋͆̇̋͒̎̈̄̓̂͊̆͂̈́̒̎͐̇̍̆̋̅̿̔͒̄̇̂̋̈́͆̎̔̇͊̊̈́̔̏͋́̀͂̈́̊͋͂̍̾̓͛̇̔̚͘̚̕̚͘͘̕̕̕̚͘͘̚̕̚̕͜͜͜͝͝͝͝͝͝͝͝ͅͅͅͅͅç̵̧̢̨̢̢̢̧̧̡̨̡̢̧̧̧̨̡̡̨̨̢̢̢̧̨̢̨̢̛̛͉̗̠͇̹̖̝͕͚͎̟̻͓̳̰̻̺̞̣͚̤͙͍͇̗̼͖͔͕͙͖̺͙̖̹̘̘̺͓̜͍̣̰̗̖̺̗̪̘̯̘͚̲͚̲̬̞̹̹͕̭͔̳̘̝̬͉̗̪͉͕̞̫͔̭̭̜͉͔̬̫͙̖̙͚͔͙͚͍̲̘͚̪̗̞̣̞̲͎͔͖̺͍͎̝͎͍̣͍̩̟͈͕̗͉̪̯͉͎͖͍̖͎̖̯̲̘̦̟̭͍͚͓͈͙̬͖̘̱̝̜̘̹̩̝̥̜͎̬͓̬͙͍͇͚̟̫͇̬̲̥̘̞̘̟̘̝̫͈̙̻͇͎̣̪̪̠̲͓͉͙͚̭̪͇̯̠̯̠͖̞̜͓̲͎͇̼̱̦͍͉͈͕͉̗̟̖̗̱̭͚͎̘͓̬͍̱͍̖̯̜̗̹̰̲̩̪͍̞̜̫̩̠͔̻̫͍͇͕̰̰̘͚͈̠̻̮͊̐̿̏̐̀̇̑̐̈͛͑͑̍̑̔̃̈́̓̈́̇̐͑̐̊̆͂̀̏͛̊̔̍̽͗͋̊̍̓̈́̏̅͌̀̽́̑͒͒̓͗̈́̎͌͂̕̚͘͘͜͜͜͜͜͠͝͝͝͝ͅͅͅͅͅͅͅS̵̡̡̧̧̨̨̡̢̡̡̡̡̧̧̡̧̢̫̯͔̼̲͉͙̱̮̭̗͖̯̤͙̜͚̰̮̝͚̥̜̞̠̤̺̝͇̻̱͙̩̲̺͍̳̤̺̖̝̳̪̻̗̮̪̖̺̹̭͍͇̗̝̱̻̳̝̖̝͎̙͉̞̱̯̙̜͇̯̻̞̱̭̗͉̰̮̞͍̫̺͙͎̙̞̯̟͓͉̹̲͖͎̼̫̩̱͇̲͓̪͉̺̞̻͎̤̥̭̺̘̻̥͇̤̖̰̘̭̳̫̙̤̻͇̪̦̭̱͎̥̟͖͕̣̤̩̟̲̭̹̦̹̣͖̖͒̈́̈́̓͗̈̄͂̈́̅̐̐̿̎̂͗̎̿̕͘͜͜͜͜͝͝ͅͅt̸̡̡̧̧̨̡̢̛̥̥̭͍̗͈̩͕͔͔̞̟͍̭͇̙̺̤͚͎͈͎͕̱͈̦͍͔͓̬͚̗̰̦͓̭̰̭̎̀̂̈́̓̒̈́̈́̂̄̋́̇̂͐͒̋̋̉͐̉̏̇͋̓̈́͐̾͋̒͒͐̊̊̀̄͆̄͆̑͆̇̊̓̚̚̕̚̕͜͠͝͝ͅͅơ̵̡̨̡̡̡̨̛̺͕̼͔̼̪̳͖͓̠̘̘̳̼͚͙͙͚̰͚͚͖̥̦̥̘̖̜̰͔̠͕̦͎̞̮͚͕͍̤̠̦͍̥̝̰̖̳̫̮̪͇̤̱̜͙͔̯͙̙̼͇̹̥̜͈̲̺̝̻̮̬̼̫̞̗̣̪̱͓̺̜̠͇͚͓̳̹̥̳̠͍̫͈̟͈̘̯̬̞͔̝͍͍̥̒̐͗͒͂͆̑̀̿̏́̀͑͗̐́̀̾̓́̌̇̒̈́̌̓͐̃̈́̒̂̀̾͂̊̀̂͐̃̄̓̔̽̒̈́̇̓͌̇̂̆̒̏̊̋͊͛͌̊̇̒̅͌̄̎̔̈́͊́̽̋̈̇̈́́͊̅͂̎̃͌͊͛͂̄̽̈́̿͐̉̽̿́́̉͆̈́̒́̂̾̄̇̌̒̈̅̍̿̐͑̓͊̈́̈̋̈́̉̍̋̊̈̀̈́̾̿̌̀̈́͌̑̍́̋̒̀̂̈́́̾̏̐̅̈̑͗͐̈͂̄̾̄̈́̍̉͑͛͗͋̈́̃̄̊́́͐̀̀̽̇̓̄̓̃͋͋̂̽̔̀̎͌̈́̈́̑̓̔̀̓͐͛͆̿̋͑͛̈́͂̅̋̅͆͗̇́̀̒́̏͒̐̍͂̓͐͐̇̂̉̑̊͑̉̋̍͊̄̀͂̎͒̔͊̃̏̕̚̕̕͘͘͘̚͘̚͘̕͘̚͘̚̚̚̕͘͜͜͜͝͝͠͠͝͝͠͠͝͝͝͝͝͝͝͝͝ͅͅͅc̴̨̡̢̢̢̡̡̢̛̛̛̻͇̝̣͉͚͎͕̻̦͖̤̖͇̪̩̤̻̭̮̙̰̖̰̳̪̱̹̳̬͖̣͙̼̙̰̻̘͇͚̺̗̩̫̞̳̼̤͔͍͉̟͕̯̺͈̤̰̹̍̋́͆̾̆̊͆͋̀͑͒̄̿̄̀̂͋̊͆́͑̑̽͊̓́̔̽̌͊̄͑͒͐̑͗̿̃̀̓̅́̿͗̈́͌̋̀̏̂͌̓́̇̀͒͋̌̌̅͋͌̆͐̀̔̒͐̊̇̿̽̀̈́̃̒̋̀̈́̃̏̂̊͗̑̊̈̇̀̌͐̈́̉̂̏͊̄͐̈̽͒̏̒̓́̌̓̅́̓̃͐͊͒̄͑̒͌̍̈́̕͘̚͘̕͘̚̕͜͝͠͝͝͝ͅǩ̴̢̢̢̧̨̢̢̢̨̨̨̢̢̢̨̧̨̡̡̢̛̛̛̛̛̛̛̜̥̩̙͕̮̪̻͈̘̯̼̰̜͚̰͖̬̳͖̣̭̼͔̲͉̭̺͚̺̟͉̝̱̲͎͉̙̥̤͚͙̬̪̜̺͙͍̱̞̭̬̩̖̤̹̤̺̦͈̰̗̰͍͇̱̤̬̬͙̙̲̙̜͖͓̙̟̙̯̪͍̺̥͔͕̝̳̹̻͇̠̣͈̰̦͓͕̩͇͈͇̖͙͍̰̲̤̞͎̟̝̝͈͖͔͖̦̮̗̬̞̞̜̬̠̹̣̣̲̮̞̤̜̤̲̙͔͕̯͔͍̤͕̣͔͙̪̫̝̣̰̬̬̭̞͔̦̟̥̣̻͉͈̮̥̦̮̦͕̤͇̺͆͆̈͗̄̀̌̔̈́̈̉̾̊̐̆̂͛̀̋́̏̀̿͒̓̈́̈́͂̽̾͗͊̋̐̓̓̀̃̊̊͑̓̈̎̇͑̆̂̉̾̾̑͊̉̃́̑͌̀̌̐̅̃̿̆̎̈́̀̒́͛̓̀̊́̋͛͒͊̆̀̃̊͋̋̾̇̒̋͂̏͗͆̂̔́̐̀́͗̅̈̋̂̎̒͊̌̉̈̈́͌̈́̔̾̊̎́͐͒̋̽̽́̾̿̚̕͘͘̚̕̕̕̚̚̕̚̕͘͜͜͜͝͠͝͝͝͝͝͝͝͝ͅͅͅͅͅͅB̸̢̧̨̡̢̧̨̡̡̨̡̨̡̡̡̢̨̢̨̛̛̛̛̛̛͉̞͚̰̭̲͈͎͕͈̦͍͈̮̪̤̻̻͉̫̱͔̞̫̦̰͈̗̯̜̩̪̲̻̖̳͖̦͎͔̮̺̬̬̼̦̠̪̤͙͍͓̜̥̙̖̫̻̜͍̻̙̖̜̹͔̗̪̜̖̼̞̣̠̫͉̯̮̤͈͎̝̪͎͇͙̦̥͙̳̫̰̪̣̱̘̤̭̱͍̦͔̖͎̺̝̰̦̱̣͙̙̤͚̲͔̘̱̜̻͔̥̻͖̭͔̜͉̺͕͙͖̜͉͕̤͚̠̩̮̟͚̗͈͙̟̞̮̬̺̻̞͔̥͉͍̦̤͓̦̻̦̯̟̰̭̝̘̩̖̝͔̳͉̗̖̱̩̩̟͙͙͛̀͐̈́̂̇͛̅̒̉̏̈́̿͐́̏̃̏̓̌̽͐̈́͛̍͗͆͛̋̔̉͂̔̂̓̌͌͋̂͆̉͑̊̎́̈́̈̂͆͑́̃̍̇̿̅̾́́̿̅̾̆̅̈́̈̓͒͌͛̃͆̋͂̏̓̅̀͂̽̂̈̈́̎̾̐͋͑̅̍̈́̑̅̄͆̓̾̈́͐̎̊͐̌̌̓͊̊̔̈́̃͗̓͊͐̌͆̓͗̓̓̾̂̽͊͗́́́̽͊͆͋͊̀̑̿̔͒̏̈́́̏͆̈́͋̒͗͂̄̇̒͐̃͑̅̍͒̎̈́̌̋́̓͂̀̇͛̋͊͆̈́̋́̍̃͒̆̕̚̚̕̕̕͘̕̚̚͘̕͜͜͜͜͝͠͠͝͠͝͝͝͝͠͝͝͝͝ͅͅͅͅͅI̵̡̢̧̨̡̢̨̡̡̢̡̧̡̢̢̢̡̢̛̛͕͎͕̩̠̹̩̺̣̳̱͈̻̮̺̟̘̩̻̫͖̟͓̩̜̙͓͇̙̱̭̰̻̫̥̗̠͍͍͚̞̘̫͉̬̫̖̖̦͖͉̖̩̩̖̤̺̥̻̝͈͎̻͓̟̹͍̲͚͙̹̟̟̯͚̳̟͕̮̻̟͈͇̩̝̼̭̯͚͕̬͇̲̲̯̰̖̙̣̝͇̠̞̙͖͎̮̬̳̥̣̺̰͔̳̳̝̩̤̦̳̞̰̩̫̟͚̱̪̘͕̫̼͉̹̹̟̮̱̤̜͚̝̠̤̖̮̯̳͖̗̹̞̜̹̭̿̏͋̒͆̔̄̃̾̓͛̾̌́̅̂͆̔͌͆͋̔̾́̈̇̐̄̑̓̂̾́̄̿̓̅̆͌̉̎̏̄͛̉͆̓̎͒͘̕̕͜͜͜͜͜͜͜͝͠ͅͅƠ̷̢̛̛̛̛̛̛̛̛̟̰͔͔͇̲̰̮̘̭̭̖̥̟̘̠̬̺̪͇̲͋͂̅̈́̍͂̽͗̾͒̇̇̒͐̍̽͊́̑̇̑̾̉̓̈̾͒̍̌̅̒̾̈́̆͌̌̾̎̽̐̅̏́̈̔͛̀̋̃͊̒̓͗͒̑͒̃͂̌̄̇̑̇͛̆̾͛̒̇̍̒̓̀̈́̄̐͂̍͊͗̎̔͌͛̂̏̉̊̎͗͊͒̂̈̽̊́̔̊̃͑̈́̑̌̋̓̅̔́́͒̄̈́̈̂͐̈̅̈̓͌̓͊́̆͌̉͐̊̉͛̓̏̓̅̈́͂̉̒̇̉̆̀̍̄̇͆͛̏̉̑̃̓͂́͋̃̆̒͋̓͊̄́̓̕̕̕̚͘͘͘̚̕̚͘̕̕͜͜͝͝͝͠͝͝͝͝͠ͅS̷̢̨̧̢̡̨̢̨̢̨̧̧̨̧͚̱̪͇̱̮̪̮̦̝͖̜͙̘̪̘̟̱͇͎̻̪͚̩͍̠̹̮͚̦̝̤͖̙͔͚̙̺̩̥̻͈̺̦͕͈̹̳̖͓̜͚̜̭͉͇͖̟͔͕̹̯̬͍̱̫̮͓̙͇̗̙̼͚̪͇̦̗̜̼̠͈̩̠͉͉̘̱̯̪̟͕̘͖̝͇̼͕̳̻̜͖̜͇̣̠̹̬̗̝͓̖͚̺̫͛̉̅̐̕͘͜͜͜͜ͅͅͅ.̶̨̢̢̨̢̨̢̛̻͙̜̼̮̝̙̣̘̗̪̜̬̳̫̙̮̣̹̥̲̥͇͈̮̟͉̰̮̪̲̗̳̰̫̙͍̦̘̠̗̥̮̹̤̼̼̩͕͉͕͇͙̯̫̩̦̟̦̹͈͔̱̝͈̤͓̻̟̮̱͖̟̹̝͉̰͊̓̏̇͂̅̀̌͑̿͆̿̿͗̽̌̈́̉̂̀̒̊̿͆̃̄͑͆̃̇͒̀͐̍̅̃̍̈́̃̕͘͜͜͝͠͠z̴̢̢̡̧̢̢̧̢̨̡̨̛̛̛̛̛̛̛̛̲͚̠̜̮̠̜̞̤̺͈̘͍̻̫͖̣̥̗̙̳͓͙̫̫͖͍͇̬̲̳̭̘̮̤̬̖̼͎̬̯̼̮͔̭̠͎͓̼̖̟͈͓̦̩̦̳̙̮̗̮̩͙͓̮̰̜͎̺̞̝̪͎̯̜͈͇̪̙͎̩͖̭̟͎̲̩͔͓͈͌́̿͐̍̓͗͑̒̈́̎͂̋͂̀͂̑͂͊͆̍͛̄̃͌͗̌́̈̊́́̅͗̉͛͌͋̂̋̇̅̔̇͊͑͆̐̇͊͋̄̈́͆̍̋̏͑̓̈́̏̀͒̂̔̄̅̇̌̀̈́̿̽̋͐̾̆͆͆̈̌̿̈́̎͌̊̓̒͐̾̇̈́̍͛̅͌̽́̏͆̉́̉̓̅́͂͛̄̆͌̈́̇͐̒̿̾͌͊͗̀͑̃̊̓̈̈́̊͒̒̏̿́͑̄̑͋̀̽̀̔̀̎̄͑̌̔́̉̐͛̓̐̅́̒̎̈͆̀̍̾̀͂̄̈́̈́̈́̑̏̈́̐̽̐́̏̂̐̔̓̉̈́͂̕̚̕͘͘̚͘̚̕̚̚̚͘̕̕̕͜͜͝͠͠͝͝͝͝͠͝͝͝͠͝͝͝͝͝͝ͅͅͅī̸̧̧̧̡̨̨̢̨̛̛̘͓̼̰̰̮̗̰͚̙̥̣͍̦̺͈̣̻͇̱͔̰͈͓͖͈̻̲̫̪̲͈̜̲̬̖̻̰̦̰͙̤̘̝̦̟͈̭̱̮̠͍̖̲͉̫͔͖͔͈̻̖̝͎̖͕͔̣͈̤̗̱̀̅̃̈́͌̿̏͋̊̇̂̀̀̒̉̄̈́͋͌̽́̈́̓̑̈̀̍͗͜͜͠͠ͅp̴̢̢̧̨̡̡̨̢̨̢̢̢̨̡̛̛͕̩͕̟̫̝͈̖̟̣̲̖̭̙͇̟̗͖͎̹͇̘̰̗̝̹̤̺͉͎̙̝̟͙͚̦͚͖̜̫̰͖̼̤̥̤̹̖͉͚̺̥̮̮̫͖͍̼̰̭̤̲͔̩̯̣͖̻͇̞̳̬͉̣̖̥̣͓̤͔̪̙͎̰̬͚̣̭̞̬͎̼͉͓̮͙͕̗̦̞̥̮̘̻͎̭̼͚͎͈͇̥̗͖̫̮̤̦͙̭͎̝͖̣̰̱̩͎̩͎̘͇̟̠̱̬͈̗͍̦̘̱̰̤̱̘̫̫̮̥͕͉̥̜̯͖̖͍̮̼̲͓̤̮͈̤͓̭̝̟̲̲̳̟̠͉̙̻͕͙̞͔̖͈̱̞͓͔̬̮͎̙̭͎̩̟̖͚̆͐̅͆̿͐̄̓̀̇̂̊̃̂̄̊̀͐̍̌̅͌̆͊̆̓́̄́̃̆͗͊́̓̀͑͐̐̇͐̍́̓̈́̓̑̈̈́̽͂́̑͒͐͋̊͊̇̇̆̑̃̈́̎͛̎̓͊͛̐̾́̀͌̐̈́͛̃̂̈̿̽̇̋̍͒̍͗̈͘̚̚͘̚͘͘͜͜͜͜͜͜͠͠͝͝ͅͅͅ☻♥■∞{╚mYÄÜXτ╕○\╚Θº£¥ΘBM@Q05♠{{↨↨▬§¶‼↕◄►☼1♦  wumbo╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ╚̯̪̣͕̙̩̦͓͚̙̱̘̝̏̆ͤ̊̅ͩ̓̏̿͆̌Θ̼̯͉ͭͦ̃͊͑̉ͯͤ̈́ͬ͐̈́͊ͤͅº͍̪͇͖̝̣̪̙̫̞̦̥ͨ̂ͧ̄̿£̺̻̹̠̯͙͇̳ͬ̃̿͑͊ͨͣ

Link to comment
Share on other sites
Link to post
Share on other sites
Posted

I've already seen what the frame generation looks like on Hardware Unboxed's video on the matter.

 

To me, frame generation matters only if we can ever get low end to midrange Ada GPUs.

We need more tests on seeing if frame generation would differ on lower settings, lower base framerates, and maybe, without motion blur.

 

Link to comment
Share on other sites
Link to post
Share on other sites
Posted
5 hours ago, Origami Cactus said:

IT GENERATES THE INTERMEDIATE FRAMES. Not a future frame.

I think Nvidia's own wording on that topic isn't super clear.

 

Does "current frame" refer to the one you're seeing right now (frame 1) and "previous frame" to the one that was visible previously (frame 0)? Does the AI predict the intermediate frame 1.5 based on frame 0 + 1 before the engine has rendered frame 2? That would result in much less added latency.

 

Or does "current frame" refer to the one the engine has generated (frame 2) and "previous frame" refer to the one that's actually visible on the monitor right now (frame 1)? Does the AI generate the intermediate frame 1.5 essentially by interpolation, holding back frame 2 until frame 1.5 was shown?

 

Nvidia also calls it "predictive frame generation" further implying the AI predicts the intermediate frame before the next frame was generated by the engine, rather than delaying engine frames.

Remember to either quote or @mention others, so they are notified of your reply

Link to comment
Share on other sites
Link to post
Share on other sites
Posted

The video seems to find that:

  • If you've got a great framerate already, you can turn this feature on and it'll... still look great.
  • But if you've got a terrible framerate, you can turn this feature on and it looks... worse.

Are we sure this isn't an audiophile product?  Because that sales pitch makes NO sense.


There's also the issue of interpolating.  Surely it isn't interpolating.  Interpolating means it generates frame A, shows it to you, generates frame C, DOES NOT SHOW IT TO YOU, then generates frame B, and show that to you while frame C gets further and further out of date.  This would imply that the latency ought to be greater than one full frame at all times.

 

If it was extrapolating frames, that'd make a lot more sense than interpolating them.   To be clear, extrapolating would mean you use frames A and C to produce D and display it while you're calculating E.  This would not dramatically increase latency.   (Although then you will occasionally show the user a completely wrong frame.)

Link to comment
Share on other sites
Link to post
Share on other sites
Posted
13 hours ago, Eigenvektor said:

I think Nvidia's own wording on that topic isn't super clear.

 

Does "current frame" refer to the one you're seeing right now (frame 1) and "previous frame" to the one that was visible previously (frame 0)? Does the AI predict the intermediate frame 1.5 based on frame 0 + 1 before the engine has rendered frame 2? That would result in much less added latency.

 

Or does "current frame" refer to the one the engine has generated (frame 2) and "previous frame" refer to the one that's actually visible on the monitor right now (frame 1)? Does the AI generate the intermediate frame 1.5 essentially by interpolation, holding back frame 2 until frame 1.5 was shown?

 

Nvidia also calls it "predictive frame generation" further implying the AI predicts the intermediate frame before the next frame was generated by the engine, rather than delaying engine frames.

It is currently the second variant. As you said, it would be much better if it was the first variant, but it is not.

 

A simple proof that it is the second variant is what happens when a scene changes in a single frame. The generated frame is a artifact filled mess that is the combination of previous scene and the new one. If it was the variant one, the predicted frame would just be the continuation of the previous scene, and then the scene would change after the predicted frame without any artifacts.

 

Prediction doesn't always have to be about the future you know. Currently Nvidia "predicts" what the frame between 2 actual frames would look like. I recommend watching the nvidia keynote about the subject, they also have a pretty well made scheme about the pipeline.

I only see your reply if you @ me.

This reply/comment was generated by AI.

Link to comment
Share on other sites
Link to post
Share on other sites
Posted
9 hours ago, Origami Cactus said:

If it was the variant one, the predicted frame would just be the continuation of the previous scene, and then the scene would change after the predicted frame without any artifacts.

I agree with what you said, but… I would assume if it predicts (extrapolates) frame 1.5 from information found in frame 0 and frame 1, there would still be artifacts if these frames were vastly different. Same as predicting (interpolating) what 1.5 should've looked liked from information in frame 1 and 2.

 

That being said, while extrapolation should get rid of the latency penalty, I think you'd run into much bigger issues with visual artifacts. If there's an object that is (partially) occluded in frame 1 and visible in frame 2, then interpolation can generate an in-between state. It only has to remove information that should not be visible in frame 1.5 yet. If instead you try to extrapolate from two previous frames, you can't generate the missing information out of thin air.

 

Just imagine turning your head. Pixels that don't come into view until frame 2 aren't visible in either frame 0 or frame 1. You now have an area at the edge of the view in frame 1.5 you need to fill with something but you have no idea what it is, because that information isn't available until frame 2 has been rendered.

 

So I imagine you'd have constant artifacts at the screen border and around every object occluding something else, especially if your view changes rapidly.

Remember to either quote or @mention others, so they are notified of your reply

Link to comment
Share on other sites
Link to post
Share on other sites
Posted
52 minutes ago, Eigenvektor said:

I agree with what you said, but… I would assume if it predicts (extrapolates) frame 1.5 from information found in frame 0 and frame 1, there would still be artifacts if these frames were vastly different. Same as predicting (interpolating) what 1.5 should've looked liked from information in frame 1 and 2.

 

That being said, while extrapolation should get rid of the latency penalty, I think you'd run into much bigger issues with visual artifacts. If there's an object that is (partially) occluded in frame 1 and visible in frame 2, then interpolation can generate an in-between state. It only has to remove information that should not be visible in frame 1.5 yet. If instead you try to extrapolate from two previous frames, you can't generate the missing information out of thin air.

 

Just imagine turning your head. Pixels that don't come into view until frame 2 aren't visible in either frame 0 or frame 1. You now have an area at the edge of the view in frame 1.5 you need to fill with something but you have no idea what it is, because that information isn't available until frame 2 has been rendered.

 

So I imagine you'd have constant artifacts at the screen border and around every object occluding something else, especially if your view changes rapidly.

Actually wouldn't be much of a problem, if you would take the approach that VR currently uses, Asynchronous SpaceWarp. Instead of rendering fake sphagetti frames, you take your head movements (or mouse movements on the pc), and you translate the previous frame around, till a newer frame is rendered. You get around the screen border artifacts by taking another VR technology, foveated rendering, and rendering a bit of the area outside the screen edge at progressively lower resolutions, and then smoothing it so it doesn't look pixelated. Yes, it takes a bit of performance hit to do that, but what you get in return is a gaming experience where your screen framerate, and the framerate the game actually runs at are separate. 2kliksphilips made a really good video about it, and I like that technology a bit more than the current DLSS 3.0 implementation. I can already see DLSS 3.1 including some of those methods discussed in that video.

I only see your reply if you @ me.

This reply/comment was generated by AI.

Link to comment
Share on other sites
Link to post
Share on other sites
Posted

Did this video have fake 60fps? Something about the fps seemed off, but maybe its because i'm not used to ltt videos at 60fps

Link to comment
Share on other sites
Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share
Go to topic listing LTT Releases

×