How DLSS 5 Uses AI to Make Games Look Real

David (0:00)

Welcome to techdaily.ai. I am your host David, and joining me today is our resident expert, Sophia.

Sophia (0:05)

Hey everyone, it's fantastic to be here.

David (0:08)

This podcast is sponsored by Stonefly, your trusted solution provider and advisor in enterprise storage, backup disaster recovery, hyper-converged VMware, Hyper-V, Proximox cluster, AI servers, SA365, a turnkey security appliance, public and private cloud. Check out stonefly.com for more info.

Sophia (0:28)

So we have a really incredible deep dive plan for you today.

David (0:31)

We absolutely do. I want you to imagine something for a second. Imagine you have exactly 16 milliseconds to paint a photorealistic masterpiece. Oh, wow.

Sophia (0:40)

Okay.

David (0:41)

Someone just hands you a canvas, the clock starts ticking, and before you can even blink your eyes, you have to hand back a perfect window-like recreation of reality. Ready? Go.

Sophia (0:50)

Which is literally impossible for a human.

David (0:51)

Completely impossible. But that impossible deadline is the fundamental roadblock the video game industry has faced for decades, really.

Sophia (0:59)

Yeah, the classic 16 millisecond problem.

David (1:01)

Exactly. And today, we are exploring a technological leap that is fundamentally altering digital reality. We're talking about Nvidia's new Dlss 5

Sophia (1:11)

And the tech world isn't just calling this some incremental update. They're calling it the greatest breakthrough in computer graphics to the debut of real time ray tracing back in 2018

David (1:22)

Right, because it's blending handcrafted rendering with generative AI.

Sophia (1:25)

Yeah, it truly is a paradigm shift. But to really understand why this is sending such shockwaves through the entire industry, we kind of have to look at the mechanics of that 16 millisecond problem you just mentioned.

David (1:36)

Let's break that down for everyone.

Sophia (1:38)

So it all comes down to what we can call the cinematic gap. Think about the visual effects in like a massive Hollywood blockbuster.

When you sit in a theater and you watch a photorealistic spaceship fly through this gorgeous alien nebula, or you watch a perfectly rendered digital creature interact with live actors, you are looking at offline rendering.

David (1:57)

Right, because those frames are processed by massive server farms for days or even weeks before anyone ever actually sees them in the theater.

Sophia (2:05)

Exactly. The computers have literally all the time in the world to calculate the physics of the scene. They run these incredibly complex mathematical equations to determine exactly how millions of individual photons of light shoot out of a star.

And then travel through the dust of a nebula, bounce off the metallic hull of the spaceship, and finally hit the virtual camera lens.

David (2:26)

It is pure unadulterated mathematical simulation.

Sophia (2:29)

But, and this is the huge, but when you are sitting on your couch playing a video game, you don't have days. A video game is interactive. When you push the joystick forward, the world has to react instantly.

David (2:40)

Right, because if it doesn't, the game feels sluggish, right? It feels totally unplayable.

Sophia (2:45)

Exactly. To hit that gold standard of 60 frames per second, which is what you really need for a game to feel truly responsive, your console or your PC has to build an entirely new frame from scratch every 16.6 milliseconds.

David (2:58)

Which is just wild when you think about it. It has to read your controller input, update the game's logic, figure out where all the characters moved.

Sophia (3:05)

And then draw millions of pixels on your screen with lighting, shadows and textures.

David (3:09)

All in 16 milliseconds.

Sophia (3:12)

And the staggering difference between having hours to mathematically simulate light for a movie and having 16 milliseconds to fake it for a game is why interactive digital worlds have never quite achieved true photorealism.

David (3:26)

But I mean, it certainly hasn't been for a lack of trying, right?

Sophia (3:29)

Oh, definitely not. The entire history of graphics hardware has basically been this massive brute force arms race to solve this exact bottleneck.

David (3:37)

For over two decades, engineers try to solve this by just building bigger and bigger engines. If we look back to 2001, Nvidia introduced programmable shaders with the GeForce 3

Sophia (3:49)

Which was huge at the time.

David (3:50)

Oh, it was massive. Before that, graphics cards were pretty rigid. They basically just slapped flat textures onto polygons. But shaders allowed developers to write custom code, telling the hardware exactly how to color and light specific surfaces.

Sophia (4:02)

It was sort of like moving from a rigid paint by numbers kit to actually being able to mix your own paints.

David (4:07)

Yeah, that's a great way to put it. Then, the real heavy lifting started in 2006 with the introduction of CUDA on the GeForce 8800 GTX.