The PowerVR architecture is 25 years old

This summer, Imagination Technologies, initially going by the name VideoLogic, is celebrating the 25th anniversary of its PowerVR architecture as well as its Tile Rendering. The company took advantage of this occasion to publish a series of posts on its blog about its history in the field of 3D technology.

The first of these posts speaks about the company’s beginnings in 1997 and the first card it developed on the basis of FPGA Altera – a company bought by Intel in 2015 – and Xilinx. Due to their lack of computational power, these cards were not able to render textures, something that the Midas 1 was capable of since it made use of dedicated ASICs, one of which was dedicated to texture rendering. Just like the first 3dfx Voodoo cards, the Midas 1 made use of a VGA pass trough; the 2D card was connected to the 3D card via VGA and the 3D card was then connected to the display. Midas 3 did away with this solution by allowing the transfer of data to the 2D card via the PCI bus.

In association with NEC, it is with the PCX1 (1996) and PCX2 (1997) that VideoLogic first began to use the name PowerVR for its chips and to become known among the general public. These 3D acceleration chips, with 500 or 350 nm etching, operated at 60-66 MHz and had a fillrate of 60-66 megapixels and a bandwidth of 0.48 to 0.528 GB/s (!!!!), the specific advantage of the PCX2 being its bileniar texture filtering support, something which was sorely lacking on the PCX1 compared to the Voodoo Graphics. These chips found their way onto 3D acceleration cards (such as the Matrox M3D, VideoLogic Apocalypse 3D) as well as on the Apocalypse 5D which also incorporated a 2D chip – by Tseng Labs – the ET6100.

At this point in time, the use of standardized APIs (OpenGL, Direct3D) in the computer entertainment field was at its very beginnings and users therefore required a specific card running the proprietary SGL API, however the number of cards that used this standard was much lower than those which adopted 3dfx’s Glide API which led to the eventual domination of this standard.

The following generation of chips was used by Sega in 1998 for its Dreamcast console, a very lucrative contract which did nothing to aid in the release of the PC version (the Neon 250, in 1999) which was outperformed by Nvidia’s Riva TNT2.

Now in association with STMicroelectronics rather than NEC, the company began work on the third generation of the PowerVR, the STG4000 (2000) and STG4500 (2001), better known by the names Kyro and Kyro II. It is in 2001, namely with the Hercules 3D Prophet 4500 which we tested, that the first cards were announced. Lacking a geometry engine (T&L), contrary to Nvidia’s GeForce 256 (1999) and ATI’s Radeon (2000), they were (successfully) geared towards users seeking a better performance to price ratio.

But when ATI and Nvidia started churning out GPUs, STMicroelectronics decided to leave the PC GPU market in 2002 (which only represented 0.2% of its revenue at the time), which spelled the end for the 4th generation of PowerVR, but also the end of PowerVR on PCs.

From then on, Imagination Technologies began to successfully concentrate on the mobile market, licensing architectures to be integrated into SoCs, namely Apple’s, beginning in 1998. But, this lucrative partnership, which represented as much as 50% of the company’s revenues will be coming to an end since Apple announced last April that it will stop using Imagination’s architectures, and will be replacing them with their own within the next 2 years. This announcement resulted in a 70% decrease in Imagination’s stock price, which is probably the reason why it was officially put up for sale in June.

Leave a Reply

Your email address will not be published. Required fields are marked *