The Next Leap: How A.I. will change the 3D industry

An interesting look into how AI is changing the 3D industry.

 

I watched 6 minutes and the only thing I learned is that Arowx real life name is Andrew Price.
Button your shirt.

 

Good video. I'd seen most of the examples before but it's a good summary of where things are at.

 

Substance Designer is not AI, you use nodes, the better textures are very complex spagheti nodes.
It will be AI when you'll drop some photos and Substance will create the materials and make the graph nodes itself alone with one button to press "Make my textures" LOL

 

"is AI" is a moving target. Every time some new tool or techniques is automated, the bar for "is AI" moves a little bit further.

 

Why not.
But ai can't do automatic substance designer graph creation from photos , like geussing the shapes, the details and make all the spaghetti code lol
This is something not available, or ai able to create best toplogy quads low poly models from one or two photos without needing photogrammetry, another one is ai create animation in some rig using a simple movie.
This is where ai would be the most helpful about content creation.

 

Actually the fascinating bit is where AI enhanced denoising can upscale lower resolution images, could this boost a games higher resolution performance?

And would you need a GPU with built in denoising hardware like RTX 2080 or could older GPUs denious and scale fast enough for a game?

 

No.

 

Also... why?

........

Just use larger textures. If you are upscaling, the memory footprint is the same, but now you are adding preprocessing to the image data and non trivial operations at that. You are trading a tiny amount of cheap memory for expensive gpu/cpu time. Though kinds of things are for pipeline not runtime.

 

I will be interested to see, when we can run 30-60 or maybe even 100fps scaling in real time, on potatoes.

 

Are you sure... Real Time denoising of medical imagery using a GPU -> https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5985814/

 

You do realise that's all the GPU is doing, right? That it's dedicating all its processing power to denoising the images, right? And that they're using multiple sets of low resolution images, right? And that it needed to be trained on loads of images from the start, right? They also don't even seem to mention what GPU they're using. Please read these things past the title and abstract.

 

Apologies did a quick google search, here is Nvidia's game scene specific Ray tracing and denoising system.

I think they talked about it more in the RTX presentation.

Imagine the performance benefits if RTX or other GPUs can upscale and denoise from 1k to 2k to 4k with faster rendering/upscaling/denoising than a higher resolution render?

 

You brought up older GPUs. The RTX series is not an older GPU. It is literally the opposite of an older GPU.

 

Medical imagery makes sense because medical imaging equipment is relatively poor. Its a super useful technique when you need to get a better quality image.

But it seems pretty useless for games when you can already start with a high quality image.

 

Modern high end cards struggle to get 60 fps @ 4k, there are bandwidth and processing limitations when rendering with all the processes needed to create a good looking scene that do not scale well e.g. Lighting, Depth Buffers, AO, Shadows that make Deferred and Forward rendering struggle at larger resolutions.

https://forum.unity.com/threads/is-unity-looking-into-4k-rendering-pipelines.433980/

 

Wouldn't be an Arowx thread without an Ashes of the Singularity benchmark post that has nothing to do with anything.