Representatives of Imagination refused to discuss whether or not the A4 SoC uses an Imagination core. But Apple owns just under 10 percent of the company and all iPhone and iPhone touch models use Imagination’s PowerVR MBC family of graphics cores. Imagination also recently confirmed that the iPhone 3GS uses the upgraded PowerVR SGX design. If the iPad continues this trend, it could take advantage of features of the Imagination graphics core that are uniquely well-suited to driving a screen as large as the one on the iPad.
For instance, Imagination uses so-called “tile-based deferred rendering,” which helps drive a faster user interface. “You split a screen into little tile zones,” says Kristof Beets, manager of business development for graphics at Imagination. This allows a chip’s graphics cores to compute individual tiles of the screen–say, 32 by 32 pixels on an 800 by 480 screen, with data stored in on-chip caches. By avoiding the step where a full-screen renderer has to access RAM, the chip can render a screen full of images much faster.
A second feature of Imagination’s technology that may be relevant is “deferred rendering.” Normally, a 3-D algorithm will compute the location data of a given object after computing its shape and the lighting effects applied to it. This means that where pixels on a screen correspond to objects that are blocked by other objects, some of that computation is wasted. The same is true for objects in windows layered one on top of the other in a desktop environment. Imagination’s chips, in contrast, compute the location data first, minimizing the number of computations that must be made and allowing for lower power consumption.
In April 2008, Apple acquired P.A. Semi, a chip manufacturer that specialized in power-efficient processors that use the PowerPC architecture–the same architecture used by Apple in its computers until it switched over to Intel CPUs in 2006.
“Some of [P.A. Semi’s] engineers had ARM experience, and, of course, their chip-design knowledge would be transferrable to any CPU architecture,” says Halfhill. “A highly integrated SoC like the Apple A4 would take at least 12 to 18 months to design, debug, and manufacture, however, making it unlikely that P.A. Semi engineers designed it from scratch.”
In Halfhill’s view, this makes it even more likely that the A4 chip is made primarily of designs that closely match existing ARM cores. “Apple would have had to move awfully fast to design its own ARM-compatible core and the A4 SoC in so short a time,” he says. “That’s why I think the A4 is built on existing cores from ARM.”
Halfhill suggests that P.A. Semi engineers may have been brought on board for some project other than the A4 chip. “I wouldn’t be surprised if many or most of the P.A. Semi engineers were assigned to another project–such as a future Apple A5 chip,” he says.