Ohhhh raytracing is required? Awesome, I can’t play it so I guess I don’t even have a reason to follow it for price drops. Oh well, Selaco’s new campaign act is about to drop, I’ll guess I will give it a rerun instead of playing doom.
There was a tech talk about Quake 3 - surprisingly not by Carmack - which highlighted how id is technologically conservative. Quake 3 was their first game with no software renderer, and it still asked for nothing but OpenGL 1.1 and a Pentium II. It somehow featured curved surfaces, volumetric fog, and believe it or not, shadow volumes. Dynamic lighting was all Blinn-Phong from 1977. Static lighting was one (1) 128x128 lightmap, for dramatic gradients all over a stadium-sized level.
Doom 3 had steeper system requirements because it still did shadow volumes on the CPU and re-rendered most of the scene for each of them. But if you look at a screenshot with lighting disabled, the polygon count was closer to Quake 2, with bump maps adding all of the detail.
Rage, with its unique texel for every square inch of its gigantic world, could both run on an original iPhone and stream from a DVD on Xbox 360.
So it’s really fucking weird to see them demand raytracing. Look: I’ve been following real-time raytracing on GPUs since 2009, when Ray Tracey on Blogspot coerced it out of his GTX 300-series. It seemed like an obvious choice, once we figured out how to use fewer rays. (Blending with past frames was an ugly kludge; obviously that wouldn’t continue.) I had mixed feelings when Nvidia made it yet another proprietary anticompetitive gimmick. I do not understand how modern cards have hardware specifically for this thing - and it still chugs. It just uses more rays. Like your low-frequency indirect lighting needs multiple samples per-pixel, instead of updating some probes.
Quake 3’s volumetric fog used naive raymarching. By the PS3 era we’d figured out you can do it badly, per-pixel, and then blur. OpenGL 1.1 didn’t do “blur.” OpenGL 1.1 barely “per-pixel.” id Software did it the hard way, in tiny steps, on the CPU. And yet it still ran great, because they did it per vertex, and blended across wobbling triangles, and it looked fucking great.
I’m tempted toward an “eat hot chip and lie” rant about modern developers who can’t imagine doing anything only a thousand times per frame.
Ohhhh raytracing is required? Awesome, I can’t play it so I guess I don’t even have a reason to follow it for price drops. Oh well, Selaco’s new campaign act is about to drop, I’ll guess I will give it a rerun instead of playing doom.
There was a tech talk about Quake 3 - surprisingly not by Carmack - which highlighted how id is technologically conservative. Quake 3 was their first game with no software renderer, and it still asked for nothing but OpenGL 1.1 and a Pentium II. It somehow featured curved surfaces, volumetric fog, and believe it or not, shadow volumes. Dynamic lighting was all Blinn-Phong from 1977. Static lighting was one (1) 128x128 lightmap, for dramatic gradients all over a stadium-sized level.
Doom 3 had steeper system requirements because it still did shadow volumes on the CPU and re-rendered most of the scene for each of them. But if you look at a screenshot with lighting disabled, the polygon count was closer to Quake 2, with bump maps adding all of the detail.
Rage, with its unique texel for every square inch of its gigantic world, could both run on an original iPhone and stream from a DVD on Xbox 360.
So it’s really fucking weird to see them demand raytracing. Look: I’ve been following real-time raytracing on GPUs since 2009, when Ray Tracey on Blogspot coerced it out of his GTX 300-series. It seemed like an obvious choice, once we figured out how to use fewer rays. (Blending with past frames was an ugly kludge; obviously that wouldn’t continue.) I had mixed feelings when Nvidia made it yet another proprietary anticompetitive gimmick. I do not understand how modern cards have hardware specifically for this thing - and it still chugs. It just uses more rays. Like your low-frequency indirect lighting needs multiple samples per-pixel, instead of updating some probes.
Quake 3’s volumetric fog used naive raymarching. By the PS3 era we’d figured out you can do it badly, per-pixel, and then blur. OpenGL 1.1 didn’t do “blur.” OpenGL 1.1 barely “per-pixel.” id Software did it the hard way, in tiny steps, on the CPU. And yet it still ran great, because they did it per vertex, and blended across wobbling triangles, and it looked fucking great.
I’m tempted toward an “eat hot chip and lie” rant about modern developers who can’t imagine doing anything only a thousand times per frame.
It’s not like the game can’t launch without a newish GPU, it’ll just run like ass.