Slashdot Mirror

The tricks played to make things look real have been very convincing and took up less power.

But go look at a teardown of a single scene in GTA V

http://www.adriancourreges.com...

"All in all there were 4155 draw calls, 1113 textures involved and 88 render targets."

And a lot of clever trickery that engine etc. programmers have to apply, texture artists have to take account of, etc. etc etc.

The "shortcuts" give convincing near-realism on low hardware for a hefty development price.

Ray-tracing gives convincing realism on high hardware for much less development all over.

As such, I can see a move towards ray-tracing.... eventually. Especially if you can put in a hardware features that helps it, wrap that in an API and force people to buy not just a GPU or a GPGPU but something that's capable of accelerating raytracing specifically. Then you could convince gamers that they need a new card. Especially if you consider, say, VR on top - VR and ray-tracing seems to go together naturally.

Now you have the same incentive as people who had to go out and by an FPU, a Voodoo card, etc. etc. to play the latest-greatest game that consoles can't get near for another few years. It's what sold Quake, which in turn sold OpenGL cards.

If things go right, VR/RT will be the reason that people upgrade their machines to play the next mega-game. And developers will push for that not just for showmanship but because their lives suddenly become cheaper and easier. Why bother to sit and texture a glass and come up with so-many-different bump maps, texture maps, reflection maps, etc. etc. when you can just specify that it's glass and have done with it and let the API convert it.

I remember playing about with POV-Ray back in the day. My old Pentium could barely render a frame in a day. But the descriptive language used to generate the scenes was much easier than the things I see being shoved into graphics memory nowadays.

If you pull the description of material properties out of the artists hands and into an API or engine, and then let the hardware do the description and heavy-lifting, ray-tracing is suddenly much easier than trying to craft it all yourself.

The same way that early 3D games all had to write their own 2.5D / 3D engines (e.g. Doom) that did everything themselves and were overtaken by a handful of OpenGL statements and compatible hardware that didn't need the clever tricks, massive optimisation and limitations.

And if you can get real-time ray-tracing, the creation tools basically mirror the production hardware. You have no need to pre-render, Z-order, edge-cull or almost anything else. You just describe the scene as you like, and that's your finished scene - the ray-tracing handles the rest.

I can quite see an era where ray-tracing takes control of the industry because it's just not worth faffing about optimising when you can just throw the whole scene at the hardware and let it do the work.

Maybe at that point, games will return to gameplay and atmosphere rather than just eye-candy, because they'll all be photo-realistic with almost no extraneous effort over downloading a library object and putting it into a scene.

The same site also posted detailed looks at Grand Theft Auto V, Supreme Commander, and Deux Ex: Human Revolution earlier this year.