I wish GPU manufacturers adopted your adaptive shading idea. It is so simple and in the same time brillliant. Like, there's no need in this blurry and not nice upscalers like FSR or DLSS. You just get several times performance boost in fragment stage for free. Wow 😮
Glad that you covered this topic of differed shading. We don't get to learn such advanced topic that easily or effectively. Hope you keep sharing the knowledge on similar topics. Cheers !! 👏👏👏
Can you use tensor cores to figure out adaptive shading across frames? DLSS fills in missing pixels for un upscaled image. Can we use a similar tehnique to fill in missing color in VRS blocks that are not 1x1? If possible that would allow to use tensor cores in the shading process, not just in the upscaling.
After watching this video I have made deffered shading for my scene, and my FPS dropped from 27 to 11. :( I suppouse not every rendering will be improved by this. The problem was that I had to give to deffered pass all textures for all possible matterials + textures with data from forward pass. And since there is a limit on number of textures per draw call (8 on laptop where I was testing), then I end up making separate draw call for each matterial for deffered pass. And I think that was to much for my 10 year old laptop.
maybe you could have read the textures during the forward pass instead? and store the albedo, normals etc in the g-buffer, so that during the deferred pass you only have to read from the g-buffer. just a thought, i haven't tried to implement it yet
You should patent your research, but you can choose who and how much you charge for a license. Big ones use free stuff but still patent their own, so why not have them pay the fee for benefiting from your time and skill.
Are you saying colleges are supposed to pander to short attention spanned mobile phone addicted teens? Uh no if you want to do hard work you have to work hard that includes working on how to make your attention span longer.
I wish GPU manufacturers adopted your adaptive shading idea. It is so simple and in the same time brillliant. Like, there's no need in this blurry and not nice upscalers like FSR or DLSS. You just get several times performance boost in fragment stage for free. Wow 😮
I've watched 6 minutes of this and I just gotta say... This dude is chill af :DD
Can't believe I've found this!! Thank you!
Great lecture! I'm glad I've discovered this gem of a channel
Probably the best explanation of the concepts I've watched so far.
You are a very inspirational researcher!
Great lecture!
Priceless!
Great visualization of complex issues! Loved the talk
"because desperation", this sum up the whole gamedev process :D
DACS is amazing!
that was a wondeful video, i will try to implement deferred shading in my opengl game engine teşekkürler (:
DACS is simple but elegant. Very cool! Thank you for haring.
I don't know what you're on man but I want some ! So chill 😀 Aside from the joke, awesome presentation 👍
Glad that you covered this topic of differed shading. We don't get to learn such advanced topic that easily or effectively.
Hope you keep sharing the knowledge on similar topics. Cheers !! 👏👏👏
Thanks a lot for sharing your knowledge . New subscriber :-)
Crysis 3, by the way, uses very thin G-Buffer (96 bits per pixel).
Doesn't the geo shader -> rasterize step create a z-buffer that will only pass the visible primitives for each pixel in forware rendering ?
Thanks a lot! Great course :)
From my experience the TAA and TXAA looks terribly on fast moving things. E.g. Aircrafts, cars, trains. It creates "ghosting".
Can you use tensor cores to figure out adaptive shading across frames?
DLSS fills in missing pixels for un upscaled image. Can we use a similar tehnique to fill in missing color in VRS blocks that are not 1x1?
If possible that would allow to use tensor cores in the shading process, not just in the upscaling.
Great lecture! Going to dig into DACS now :D
wonderfully explained
Why Forward more populer， Does Forward have ztest zwrite before Rasterizer？
I mean that Deferred would be more useful with low res and use the AI or Graphic power get the final picture？
you re the best man
I eingested all ur videos in a day
After watching this video I have made deffered shading for my scene, and my FPS dropped from 27 to 11. :(
I suppouse not every rendering will be improved by this.
The problem was that I had to give to deffered pass all textures for all possible matterials + textures with data from forward pass.
And since there is a limit on number of textures per draw call (8 on laptop where I was testing), then I end up making separate draw call for each matterial for deffered pass.
And I think that was to much for my 10 year old laptop.
maybe you could have read the textures during the forward pass instead? and store the albedo, normals etc in the g-buffer, so that during the deferred pass you only have to read from the g-buffer. just a thought, i haven't tried to implement it yet
Could you also use DACS for antialiasing, to supersample in high-frequency areas?
In theory, yes.
@@cem_yuksel DACS++?
Very cool!
You should patent your research, but you can choose who and how much you charge for a license. Big ones use free stuff but still patent their own, so why not have them pay the fee for benefiting from your time and skill.
I am very sad that my university has such a bad computer science program
Could you divide a lecture into three or four videos, because seldom of people keep focus on a serious academic video in a whole one hour.
Can't you just pause and come back to it? UA-cam will remember where you left off.
I have severely short attention span, but these lectures and the lecturer are the few that keep me engaged for a whole hour!
Are you saying colleges are supposed to pander to short attention spanned mobile phone addicted teens?
Uh no if you want to do hard work you have to work hard that includes working on how to make your attention span longer.