A History Of Mobile GPUs
In 2011, the dominant players in mobile graphics were PowerVR and Qualcomm’s Adreno. Both trace their roots back to failed x86 gaming products.
PowerVR is the most prominent contender, as Intel, Apple, and even TI use its technology. Back in the 90s, Imagination Technologies (then known as VideoLogic) started working on an infinite planes, deferred renderer graphics chip with the assistance of NEC. The PowerVR architecture was the first consumer deferred renderer in which visible pixels were drawn and occluded, or covered pixels were thrown away. At that time, other graphics chips were drawing everything, even if the person on the other side of the screen would never see the rendered output. Thus, PowerVR made much better use of available memory bandwidth, and its effective fill rate was higher, too.
The problem with PowerVR was that its team was built by visionary mathematicians and engineers without experience in chip-building and gaming. The original PowerVR PCX chip lacked bilinear filtering, which meant that buyers of the $300 graphics card saw the same pixelated textures as anyone gaming on the original PlayStation, rather than the smooth, bilinear-filtered images associated with the Nintendo 64 and 3Dfx graphics cards from the same time period. It wasn’t that bilinear filtering was a particularly challenging concept; it just wasn’t something that the engineers thought to include when they developed the chip. VideoLogic quickly came back with the PowerVR PCX2, which had a higher clock rate and bilinear filtering. Unfortunately, the PowerVR team wasn’t run by people with game development expertise. And as a result, they did not anticipate the need for src*dst texture blending. This was required for colored lighting—basically the effect needed for awesome explosions, laser beams, and alien-looking hallways. Again, that wasn’t any sort of technical challenge, but rather an issue of just not thinking about the need for this texture blending mode.
Everything was supposed to change with PowerVR Series 2, the platform used for Sega’s Dreamcast. A PC equivalent could have been the most popular graphics chip in the industry. Unfortunately, VideoLogic ran into problem after problem with its chip design. It’d tape-out and get prototypes back, only to discover a fatal glitch somewhere. One of the last problems had to do with the Windows hardware mouse cursor. Again, that wasn’t an engineering challenge, but a mistake nonetheless. The failure of PowerVR Series 2 in the PC world was ultimately what caused the company to exit the high-performance market and focus on low-power designs. It did have a short run of Series 3 chips, which lacked a hardware transformation and lighting engine, and it saw some financial success by powering digital poker machines in casinos, where the magic of deferred rendering was immediate.
PowerVR then switched from a true graphics chip manufacturer to a designer, selling its efforts the same way as ARM. This was the smartest move it could have made because it meant concentrating on core competencies based on math and block design, rather than making sure the logic was laid out ideally for a physical product. Since the company’s original architecture from the late 90s was already engineered for multi-chip design, it has been easy for PowerVR to continue to grow and evolve into the superb platform it offers today.