A Glimpse of the Graphics Grail

More papers and patents have been written and awarded on ray tracing than any other computer graphic technique.

Ray tracing is computationally burdensome. Think about just one high definition screen with two million pixels, each one needing fifty to a hundred floating point calculations every 33 ms. GPU acceleration can’t help much because one of the characteristics of ray tracing is it has no memory, no savings account, and every frame is a new frame, so the computational load is immutable.

Extraordinarily clever tricks have been developed which can, and do, deliver results that are near to or like ray tracing. Various mapping and rasterizing techniques are used, and when there are high action scenes, the viewer doesn’t have time to stop and look for breaks, baked reflections, or jaggies, so the CG developers can get away with a lot in the name of speed.

One technique that has been used successfully has been cascading shadow maps. This is similar to mip-mapping of textures. The problem with it is there are sharp edges, jumps, as you move from one map at say high resolution, to the next at a slightly lower resolution. Blending techniques and anisotropic filtering procedures have been successfully used to overcome those breaks, but to the trained critical eye, they can still be seen. Also, they don’t save much in power consumption.

Well, here they come to save the day.

Imagination Technologies has been refining the original algorithms of Caustic, a startup it acquired, to get them to fit nicely with in the silicon and computational budget needed for a low power solution.  The company now thinks they’ve done it, and to prove the point, they’ve built some actual test chips. And to make the point, they offered the comparison photograph below.

Imagination calls their architecture, Wizard. And they go so far to claim that one of the many advantages of PowerVR Wizard ray tracing GPUs lies in the power efficiency gains delivered by the architecture. The company says that the majority of data can remain in a subset of the local cache (a small savings account), eliminating the need for any high bandwidth random access path between processing elements and internal cache memory.

This coherence gathering, says the company, results in huge savings in memory bandwidth and power consumption, making Wizard ray tracing GPUs 100x more efficient compared to using GPU compute or other software-only ray tracing approaches on traditional graphics architectures. The approach has the added benefit of recollecting SIMD coherence, enabling graphics processors to tackle previously unapproachable challenges.

Those are mighty big claims, especially since ray tracing breaks a SIMD structure almost immediately due to unpredictable branches, and you can only do so much in post.

But Imagination is being clever about ray tracing. They are not ray tracing the whole scene, just the shadows, in what they call a hybrid approach.

Using their technique certain effects such as shadows, reflections and refractions are handled much better in terms of performance and quality by the ray-tracing unit while other elements in the scene can be rasterized, therefore offering developers the best of both worlds.

Overall, Imagination claims they get an average of 50% reduction in memory traffic using ray traced shadows (see its figures below).

Looking at total cycle counts, the picture is even better, says the company. They say they are seeing an impressive speed boost from the ray traced shadows. Because the different rendering passes are pipelined in both apps (i.e. the ray traced shadows app and the cascaded shadow maps app) we are unable to separate how many clocks are used for which pass. This is because portions of the GPU are busy executing work for multiple passes at the same time

That’s about as much as Imagination is saying for now. And, don’t expect to see any demos at Siggraph because they won’t be participating this year. However, Imagination has been around for many years, and they have a reputation for telling it straight, so if they say they can do this, then I’m inclined to believe them. I’ve had long stimulating conversations with the Caustic team, and I always felt like they got it, no smoke and mirrors were involved.

So we’ll have to wait to see if it works, but the wait shouldn’t be too long.

— Jon Peddie is president of Jon Peddie Research.