People are sleeping on this, so wanted to bring it to everyone’s attention. Machine Learning is a BIG deal. It will radically change how games are made and designed. Right now, you can type a description of a webpage into an algorithm and it will create it for you on the spot. You can type a description of a simple app and it can program it for you. Yes, you read that correctly.
Texture upscaling in XSX using ML has the potential to be a game changer in terms of I/O utilization. First party studios at XGS are already testing the option of shipping games with purposely lower res textures and then using ML hardware in XSX to upscale them at runtime (small DL size, small SSD footprint, drastic reduction in I/O needs, way more assets stored in RAM…with no sacrifice to fidelity at all).
That is just the start though. ML algorithms can animate characters, create textures from scratch, create photorealistic characters, even create photorealistic video sequences. I can’t even begin to scratch the surface here, but this YT channel has been highlighting this kinda stuff for yrs now: https://www.youtube.com/c/KárolyZsolnai
Imagine what game designers will be able to do in 5 yrs. Imagine verbally describing a game mechanic or scene, having one algorithm interpret your speech to text, another one translate that into computer code, yet another generating photorealistic art assets and animation sequences, and another to bug test it all.
This is a great topic! I had actually pasted some things together from the Game Stack conference that I’ll add here later. I feel like the evolution of machine learning is one of the most interesting things in the gaming industry right now. I’m more interested in this than teraflops or SSDs. Of course with a data and anaytics background, I should be. It could result in new types of game design, make games more accessible and finally, improve development efficiency. Hoping Microsoft is willing to go into more detail on this soon.
This is pretty fascinating. Is there any demos or anything normal little can try out?
Is it Elon Musk that has that brain reading tech as well. Imagine a horror game where it’s reading your thoughts (I have no idea as to what level) then ML is creating your worst nightmare on the fly. Different for everyone.
ML will effect a lot of gaming in the future as it’s implications change the way the entire stack will be used. But it’s still just learning at this point, desicision making will be the next big step. Learning is good and well, but gotta figure out what’s the best decision to make and that’s the next big step. It will take a while for ML to effect gaming in a broad sense but there will be some tangingle benefits that we will see soon.
Seeing the progress between DLSS 1 and 2 just makes fully agreeing that ML is literally the future, in terms of performance, optimization… literally everything.
I’m imagining a not-at-all-distant future where Quixel scan data can be used to train neural networks to create vast amount of photoreal art assets (SpeedTree times a billion, basically). The Flight Sim tech there looks like it could help generate rough 3D models for vast numbers of buildings and other structures too, and I bet other ML algorithms could make these photoreal even up close.
ML and procedural content creation will change a lot in the dev process. At runtime? Im not convinced the next gen consoles have enough processing power for DLSS or similar stuff.
DLSS is amazing, but runtime texture upscale is a game changer for download sizes, storage footprints, I/O needs and RAM utilization. Plus, BC games can potentially benefit from the tech too.
