More than a decade ago, old-school game coders urged their peers to reject the one true way of a fixed graphics pipeline, as embodied by the combination of Microsoft’s emerging DirectX API and the first wave of bolt-on PC graphical processing units (GPUs). The old school lost, which was initially good news for gamers. The semi-standardisation of the graphics pipeline underwrote a leap forward  in the visual quality of first PC games and then consoles, from Xbox to PlayStation 3. Technically competent 3D visuals went from being a unique selling point to the norm.

“The speed of evolution of graphics hardware has been surprisingly fast, and in many cases it seems as if it has gone faster than Moore’s Law would have predicted,” says Chris Kingsley, CTO at  Rebellion. “Graphics are an almost perfect area for parallelisation. This has allowed graphics hardware manufacturers to scale much more aggressively to massive numbers of multiple parallel execution units than CPU manufacturers have been able to achieve.”

From the start, however, developers and eye-candy enthusiasts bemoaned the sameness that a fixed pipeline imposed on game engines and their output. In the early days you could usually tell what GPU a PC game was running on simply by the graphical effects, irrespective of the title or developer. Change began five years ago with the move from fixed-function GPUs to a new generation that enabled semi-programmability through shaders. Game developers embraced the relative freedom. And now they want more.



“The DirectX pipeline has certainly stagnated,” says Tim Sweeney, CEO and technical director at Epic Games. “There’s only so much you can bolt on to the 25-year-old SGI rendering model before getting into diminishing returns territory, and that’s exactly where we are.”

Sweeney predicts that teraflop-class computing chips like Intel’s Larrabee and the innards of the next console generation will herald what he calls a “Cambrian explosion” of new graphics techniques, referring to the sudden appearance in the fossil record of most of today’s complex lifeforms.

The implication is for a new epoch in game graphics. But for Sweeney it’s more a case of backing games out of an evolutionary dead-end and “continuing the software-powered trajectory that was cut off by the advent of the first fixed-function GPUs in the late ’90s.”

Like any decent revolution, this one has no clear leader and many potential outcomes. But there is a uniting cry: freedom through programming. “The coming convergence of CPU and GPU is going to mean that graphic pipelines become even more programmable and flexible,” says Kenny Mitchell, principal programmer at Disney’s Black Rock Studio.



Dr Chris Doran, COO at lighting middleware company Geomerics, agrees. “The biggest change to the pipeline is going to come from new hardware models,” he says. “We are moving away from a fairly rigid pipeline enforced by the GPU to something more general-purpose and programmable. The  compute shader is a step in this direction, and Larrabee offers almost total freedom to customise the pipeline.”

Unlike traditional GPUs, Intel’s multi-core CPU/GPU Larrabee chip incorporates little in the way of fixed-function graphics hardware, instead supporting the traditional 3D rasterisation pipeline – or alternatives – through software. Future GPUs from rivals Nvidia and AMD are also expected to follow a similar approach, which could mean much more varied approaches to rendering game scenes in the future.

With hundreds of developers innovating in different directions, some expect to see solutions to long-standing game graphics problems, such as order-independent transparency with correct  atmospheric scattering, artefact-free soft shadows from many light sources, dynamic global illumination and more. There’s even talk of re-opening pathways dismissed as unfeasible for games, including point clouds, voxels and ray-tracing.