21 comments

[ 3.4 ms ] story [ 46.4 ms ] thread
(comment deleted)
This is awesome. It also brought back some anxiety from >10 years ago in college that reminds me that computer graphics and my brain do not agree whatsoever.
The thought expressed in the title came to my mind when I saw Nvidia described as an "AI company" in the press recently...
Do most GPUs made for AI even have a graphical output buffer and a video output any more?
Naming may provide useful hints about some utility of a tool but naming does not bound the utility of a tool.
Then who actually delivers on that front aside from AMD? Intel does deliver but only on the low to mid range.
> And here are my clearly unimpressed “friends” >:(

These friends don't get it!

Room for new competitors then? Surely Nvidia/AMD/Intel are not the only graphics vendors? Or is the tech too hard to even enter the market?
Holy tangents, Batman! This whole post was a million interrelated topics woven into one semi-coherent textbook.
Nice texture generator came out of this, with seems to be perfectly looped images! Well done!
Since when a name dictates function?
I read somewhere that CPUs are better at generating graphics than GPUs (although I imagine much slower). Is that true? Does that explain why GUI libraries like Egui are so much uglier than, for example, Iced?
They used to refer to it as GPGPU (general purpose) but they just shortened it maybe 10 years ago?
Should change the name to Matrix Processing Units
What does Bechara means? DuckDuckGo doesn't help me much. "Poor thing" maybe?

Anyway, I think you should change friends. Or study subjects. Either way.

Btw, these look like parasites or worms under microscope.

I wonder if they'll ever start doing the graphics with AI.

Could you render the scene in the simplest most efficient way possible, then train a special model that takes that picture, along with the coordinates of lights, depth maps, text descriptions of materials, etc, and adds detail?

Isnt this DLSS and Frame Gen in a nutshell?

DLSS - Render at low resolution, use an ML model to upscale

Frame Gen - Render a frame, use an ML model to generate the next frame. Render the next, gen the next....so on

(I think in practice, Frame Gen is more complicated due to fluctuation between FPS but that's another can of worms)

They're pretty much just upscaling and interpolating right? They can't fill in anything that's not actually there, like you could with a model that understands "Hey this tree is a actually just a PS2 era low poly model, but here's a picture of what it should look like, make it fit somehow"
The G now stands for GenAI.