What's your game?

Carmack Keynote at the Game Developer Conference

I just heard the Carmack keynote at GDC. Unsurprisingly, it was very interesting, but there wasn’t a whole lot of new information about Doom and what id is working on next. Note: I’m low on laptop battery, so I’m going to write now and edit later. Sorry for any typos or grammatical errors.

He started off by comparing hardware when he started programming with the hardware that’s available now. A modern system with a fast CPU and a next-gen GPU is about 1,000,000 times more powerful than, say, an Apple ][. This was a launching point to talk about how the bottleneck in 3D games has shifted in the last few years.

In the old days, actually getting the game to work took forever. From the git-go to drawing the first triangles could take several months, and drawing the first scene could take 18 months.

Now the renderers and game engine are actually the ‘easy part’. Now it takes man-months to create the actual art—that is the character models, game levels, and even textures—used in the game.

Look at character models. Some of the models in the Unreal 3 tech demo that Epic showed today are created from high-resolution models with more than 6 million polygons. It takes a very long time to create a model made up of that many polygons.

OK, uber-high polycount models and world geometry take a long time to create, but the art creation tools are much better now than before as well. There’s got to be something else that explains why it takes four years to develop a game, right?

According to Carmack, the problem is that the level of detail in games has gone up exponentially in the last five years. We’ve moved from boxy shaped objects that represent tables, to table shaped objects with realistic textures, to table shaped objects with a few objects on them to represent clutter. Next up, table-like objects that actually look like the desks we all work off of every day. Carmack says that when we can render that level of detail real-time, then we’ll basically be done with graphics in gaming from a tool and engine standpoint.

He then talked about the other technologies that need improvement, and where they stand. According to him, if we focused GPU-level hardware at audio, we’d be able to realistically model head acoustics, occlusion, reflection, and generate completely realistic audio in games.

He said the areas that really need some focus are AI and physics. AI, especially with regard to character interaction is a real problem right now. We have these extremely realistic characters, but they’re essentially cardboard cutouts with either only a few fixed lines of dialog or no real chance for character interaction.

He said physics right now is like graphics in 1997. Right now we can kind of simulate a whole bunch of things, but that there are many ways that the gamer can break the physics model, which creates a jarring anti-immersive experience. I’ll talk more about this later.

He talked briefly about a Quake2 Remix project he wanted to do where they took the existing levels and story for Quake2 but replaced the renderer and engine with something more Doom3-like. Very interesting, and the first I’d heard of that.

He mentioned that hardware curved surface rendering might be of interest as Hollywood 3D rendering and real-time 3D rendering converge. This is new, since he’d never really mentioned curved surfaces in a positive way.

He also answered a question about multi-processor console architectures by saying basically that he didn’t understand why people continued to setup their consoles with that kind of architecture, since it’s been a pain in the ass to code for for two generations. Hrmm…. I wonder if we’re going to see the next id engine debut on the PS3?

I’ll post some more cogent thoughts later.