• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Goodbye rasterization, hello NVIDIA neural rendering

erek

8=D
2FA
Joined
Dec 19, 2005
Messages
17,879
“You've got neural shaders to "unlock new compression and approximation techniques for next-generation asset generation", neural texture compression for "8x VRAM improvement at similar visual fidelity to traditional block compression at runtime", and neural materials to "compress shader code of complex multi-layered materials for up to 8X faster material processing to bring real-time performance to film-quality assets."”

Source: https://www.pcgamer.com/hardware/gr...d-were-now-firmly-in-the-era-of-ai-rendering/
 
Eventually new games will just be an empty box with a "prompt" to get the GPU going with regards to generating frames. :)
 
AI generated pixels of any kind is a tradeoff for speedup. It's just a matter of if it's worth the tradeoff. The worry is that down the road we will all forget what rasterization looked like and we'll get comfortable with what AI feeds us. Without a reference to compare, we'll never know. But ignorance is bliss, right?

When you think about it, AI the past several years has pretty much kept Moore's Law alive and well, at least in the GPU space.
 
It's just a ruse to sell AI accelerators to game dev companies, to make lazy game devs even lazier and thus games even less optimized necessitating gamers to get the latest HW. It would be brilliant if it wasn't diabolical.
 
the ai needs a good base raster work with and "similar visual fidelity" isnt up to snuff for some of us... plus theres that^^^^
 
Eventually new games will just be an empty box with a "prompt" to get the GPU going with regards to generating frames. :)
1736359985690.png
 
AI generated pixels of any kind is a tradeoff for speedup. It's just a matter of if it's worth the tradeoff. The worry is that down the road we will all forget what rasterization looked like and we'll get comfortable with what AI feeds us. Without a reference to compare, we'll never know. But ignorance is bliss, right?

When you think about it, AI the past several years has pretty much kept Moore's Law alive and well, at least in the GPU space.
While that is true Microsoft is updating DX12 to incorporate it at a low level, what we could see with future game engine design is the raster component being little more than a frame generation with placement markers, similar to how movies would green screen for CGI.
So instead of skies and mountains in the background it might just be just the outlines of the mountain and the sky with some grey scale to indicate intended lighting that the AI engine superimposes the desired effects over.
So instead of DLSS working off of Frames 1 and 5 to generate 2-4, it would just be doing all of them based on some super simple rendered outlines of a scene, and because the scene itself is so simple that could be running at hundreds of FPS and the AI subsystem is no longer generating frames based on vectors and what not but instead of actual item placement.
 
Better link: https://wccftech.com/nvidia-rtx-kit...ng-will-power-the-next-visual-leap-in-gaming/

... and suddenly the "low vram" amounts look like they'll be just fine anyway. I had no concerns, but for those that did for upcoming games.
The problem with all this stuff... is the first AAA title that releases that doesn't use it, or doesn't use it the way consumers expect will be an issue. "Game optimization" "Crappy Developers not optimizing their games". You know we are going to hear it at some point. I don't blame anyone for being upset when a $1k+ GPU has to flip ultra settings off within 2 years of its launch. 16GB I don't know its on the line imo.

Everyone would be a bit more comfortable with a belt and suspenders type fix. Up the ram, as well as adding new compression stuff. I would prefer software solutions like this be used by devs to UP texture quality in specific games by having a RTX crazy texture setting, rather then the GPU relying on game developers adopting it everywhere to justify just enough ram SKUs.
 
Last edited:
The problem with all this stuff... is the first AAA title that releases that doesn't use it, or doesn't use it the way consumers expect will be an issue. "Game optimization" "Crappy Developers not optimizing their games". You know we are going to hear it at some point.

This is cool tech and could be a great solution, I think everyone would be a bit more comfortable with a belt and suspenders type fix. Up the ram, as well as adding new compression stuff.
That would be great, but it would hinder adoption of their new compression tech.

I could go on a long rant but instead I will summarize it with this:
Nvidia is creating a problem so they can sell it's solution, if Nvidia gave the cards more ram across the board then they are pushing a solution to a problem that doesn't exist, development studios won't take the extra time to implement that tech unless it addresses a significant problem.

Side note on this, implementing this could do a lot towards solving one of the largest issues with UE5 and their outdoor "zone" loading stuttering issues.
 
While that is true Microsoft is updating DX12 to incorporate it at a low level, what we could see with future game engine design is the raster component being little more than a frame generation with placement markers, similar to how movies would green screen for CGI.
So instead of skies and mountains in the background it might just be just the outlines of the mountain and the sky with some grey scale to indicate intended lighting that the AI engine superimposes the desired effects over.
So instead of DLSS working off of Frames 1 and 5 to generate 2-4, it would just be doing all of them based on some super simple rendered outlines of a scene, and because the scene itself is so simple that could be running at hundreds of FPS and the AI subsystem is no longer generating frames based on vectors and what not but instead of actual item placement.
Fixed. ☹️
So instead of skies and mountains in the background it might just be just the outlines of the mountain Coca Cola and the sky other advertisements.
 
That would be great, but it would hinder adoption of their new compression tech.

I could go on a long rant but instead I will summarize it with this:
Nvidia is creating a problem so they can sell it's solution, if Nvidia gave the cards more ram across the board then they are pushing a solution to a problem that doesn't exist, development studios won't take the extra time to implement that tech unless it addresses a significant problem.

Side note on this, implementing this could do a lot towards solving one of the largest issues with UE5 and their outdoor "zone" loading stuttering issues.

I can see your argument. Not sure it makes me feel better spending a grand on a GPU with 16gb in '25. At least they gave the 90 32gb. Which you could also argue also makes your point not that big a deal. All the game companies really have to say is Ultra/Crazy texture mode is for x90 class cards. 80 class cards should be using High.
I mean what are you going to do the 80 is a perfectly cut in half 90 this gen its not like they could really put more then 16 gb on it anyway.
 
I can see your argument. Not sure it makes me feel better spending a grand on a GPU with 16gb in '25. At least they gave the 90 32gb. Which you could also argue also makes your point not that big a deal. All the game companies really have to say is Ultra/Crazy texture mode is for x90 class cards. 80 class cards should be using High.
I mean what are you going to do the 80 is a perfectly cut in half 90 this gen its not like they could really put more then 16 gb on it anyway.
Yeah no it's dirty pool, Nvidia is strong arming the market to either up spend to avoid the problem short term or lower settings and wait for the technology that addresses the issue to trickle down.

The only thing it does do it keep older Nvidia cards relevant longer, which ultimately only hurts AMD and Intel, as used Nvidia cards will possibly maintain their usefulness longer in the secondary markets which cuts into some of AMD and Intel's biggest customer base.

GDDR7 3GB chips will address the memory issue in time, but we just have to wait for it.
There are some arguments for and against clamshell mode with GDDR7, supposedly it introduces some added latency which slow things down enough to be noticeable.
 
  • Like
Reactions: ChadD
like this
Man it sure is nice to buy current products based on "future tech" isn't it? After much research it appears the numbers arent as strong as they initially appear, because they were surely not pure rasterization performance. No doubt it will be faster, but not what some people may have thought who just glanced at slides. Frame gen will be a big factor to get near those numbers.
 
Man it sure is nice to buy current products based on "future tech" isn't it? After much research it appears the numbers arent as strong as they initially appear, because they were surely not pure rasterization performance. No doubt it will be faster, but not what some people may have thought who just glanced at slides. Frame gen will be a big factor to get near those numbers.
But as more titles become Ray Tracing required, and fewer studios bother with any rasterization optimization processes, expect raster requirements to decrease over time not increase.
I know there are lots of people out there who are all "F Raytracing I'm leaving it off!!!" but expect those settings to look more and more potato like as time progresses.
 
But as more titles become Ray Tracing required, and fewer studios bother with any rasterization optimization processes, expect raster requirements to decrease over time not increase.
I know there are lots of people out there who are all "F Raytracing I'm leaving it off!!!" but expect those settings to look more and more potato like as time progresses.
I've never been sold on it. Might as well be hairworks 2.0 for me and many others. The cost vs benefit since it was announced really hasn't made sense. I should say it like this, if a title does not support it and or my hardware doesn't properly support it, I am not that bothered and I'd still play it.
 
“You've got neural shaders to "unlock new compression and approximation techniques for next-generation asset generation", neural texture compression for "8x VRAM improvement at similar visual fidelity to traditional block compression at runtime", and neural materials to "compress shader code of complex multi-layered materials for up to 8X faster material processing to bring real-time performance to film-quality assets."”

Source: https://www.pcgamer.com/hardware/gr...d-were-now-firmly-in-the-era-of-ai-rendering/

So essentially, lower quality. I do understand the efficiency gains can be great but if the result is still inferior that is a major downside. Seems like a lot of the technologies are just creating inferior end results.

Also this will take a while to show up in games. By the time developers start using most of these tools, we'll be looking at the next round of GPU benchmarks.
 
. I do understand the efficiency gains can be great but if the result is still inferior that is a major downside.
Here not sure how the end result would be inferior.

If you can have a scene with texture that would take 4 more time vram visually (for the texture parts), this is an superior result.
 
So essentially, lower quality. I do understand the efficiency gains can be great but if the result is still inferior that is a major downside. Seems like a lot of the technologies are just creating inferior end results.

Also this will take a while to show up in games. By the time developers start using most of these tools, we'll be looking at the next round of GPU benchmarks.
its basically cost to performance. Company gets to pump the performance number. its almost same, eye cant tell. But hey we are still going to charge you for it lol. I would expect stuff to get cheaper if its mostly software down the road but how would they make more and more profit then?
 
So essentially, lower quality. I do understand the efficiency gains can be great but if the result is still inferior that is a major downside. Seems like a lot of the technologies are just creating inferior end results.

Also this will take a while to show up in games. By the time developers start using most of these tools, we'll be looking at the next round of GPU benchmarks.
Actually the opposite, the neural compression tech Nvidia started demo'ing last year looked better and kept more details compared the the existing methods while using significantly less space.
But the algorithms are very difficult to do on traditional GPU's, and makes extensive use of the tensor cores.
https://research.nvidia.com/labs/rtr/neural_texture_compression/assets/ntc_medium_size.pdf

Obviously raw is best, but games aren't giving us raw textures.
So in most of the cases they have shown off since releasing the technology, they have routinely managed equal or better quality at significantly smaller file sizes.

From the paper here.
1736380700966.png


Most would be hard pressed to tell the difference between the BC Medium (what most games use for 4K) and NTC 1.0 (Nvidia's AI Compression), but that file size difference is huge.

And at lower resolutions like 1080p the NTC 0.2 option looks infinitely better than the BC Medium option aiming for a similar file size.
 
Back
Top