GPU Designer Plans To Stuff A PS3 In Your Pants
This sort of capability is just one of Imagination Technologies' targets. The company has already been working with Adobe to develop PowerVR-compatible Flash acceleration functionality and already supports OpenCL. Going forward, PowerVR intends to explore using multiple GPUs in concert (think mobile SLI); a process which could theoretically boost performance considerably without drawing additional power or generating more heat than a handheld device can dissipate.
With the iPhone already capable of this... We're off to a pretty good start
Claiming that a console will fit into a pocket in three years is a bold statement but not necessarily an impossible one. Let's take a look.
NVIDIA And PowerVR: Past, Present, And the PS3
The PS3's GPU is based on NVIDIA's G70 (NV47) architecture, first unveiled in June 2005. It was the last NVIDIA design to feature fixed pixel and vertex shader pipelines (24 PS, 8 VS) and it's not programmable in the same sense that a modern G80/R600-derived solution is. Even if it's not a modern GPU, the Reality Synthesizer inside the PS3 is several orders of magnitude more powerful than your current PowerVR SGX535 processor and considerably 'wider.' The real-world size and heat constraints of a modern handset make PowerVR's claims sound like hype, and they might be, if PowerVR used standard rendering techniques. They don't.
The Hercules Prophet II 4500. Definitely a unique solution in its day, but no, it can't play Crysis.
Nine years ago, in the chaotic rearrangement of products and preferences that followed 3DFX's sudden collapse, PowerVR partnered with STMicroelectronics and brought a GPU to market called the Kyro II. On paper, the Kyro II looked like a joke against NVIDIA's GeForce2 GTS; it had less than half of that card's theoretical fillrate, half its memory bandwidth, just two pixel pipelines (compared to four on the GF2) and couldn't apply more than one texture in a single pass.
What made the Kyro II competitive was the fact that it used tiled rendering; we can explain the difference using the Half Life 2 screenshot above. Looking at the image, you can tell that some barrels and objects are in front of and partially obscure others. Asked to render the above scene, the GeForce 2 would draw all of the objects completely, including those objects we can only see part of (or maybe can't see at all). This wasn't a problem at low resolutions or 16-bit color but proved staggeringly inefficient as games became more visually complex.
Kyro II sidestepped this problem by breaking the image into tiles and only rendering what was actually visible. As a result, it scaled much more effectively than any other GPU on the market as detail levels, resolutions, and color depth were raised. Ironically, the PowerVR solution actually competed more effectively at higher resolutions than lower ones—for a little while it seemed as though PowerVR might take 3DFX's place and maintain a three-way competition in the GPU market. This didn't happen for a variety of reasons, no Kyro III ever materialized, and PowerVR faded off the radar.
Nine years later the company is still using tiled rendering and the inherent efficiency of the approach could make it an ideal solution for mobile devices where low power consumption and small die size are more important than theoretical peak performance. Perhaps most interesting is the fact that PowerVR's 3D solutions three years hence will almost certainly be bumping up against NVIDIA's Tegra. We might all have front-row seats to a grudge match nearly 15 years in the making.
We'll bring popcorn.