Graphics chip developer Nvidia will target smart phone manufacturers next year with the launch of a product combining its GPU with the application processor technology it got last year with its acquisition of PortalPlayer.
We’ll launch the product in January, competing in the smart phone arena with app processor vendors TI, Samsung, and Broadcom, said Mike Hara, VP of corporate comms and investors relations at the Santa Clara, California-based vendor.
The smart phone market, just like PCs, is evolving into two segments: the feature phones with the baseband, app, and graphics processing all going on in a single processor, and the smart phone space, where they are done separately, said Hara. Today smart phones are running their graphics on the app processor, which results in poor battery life, but our product will address that with a dedicated GPU.
The other big area of development at Nvidia is in high-performance computing, where the company said the GPU’s greater clout in floating-point operations compared to regular CPUs stands it in good stead to compete for business currently going to x86 clusters.
The Tesla chip launched earlier this year for this market does 500 gigaflops, whereas a Core 2 Microprocessor [from Intel] does 10, said Hara. He said the HPC market is a $10bn market, served entirely by CPUs, so Nvidia is targeting any floating-point-intensive market within HPC. Today that means oil and gas, medical imaging, and the financial sector, but Nvidia intends to add new verticals.
Our Cuda development environment, which consists of a C compiler, a profiler and an optimization tool, has been available to download from our web site since June, said Hara. We’ve had 30,000 downloads and have been able to identify 3,600 developers in 14 different industry categories.
Nvidia is not alone among GPU vendors in seeking to address the HPC market. The other dedicated GPU player is ATI, which was acquired last year by AMD. Its offering in this space is called FireStream. Hara said a major differentiator for Nvidia is the fact that the Tesla is programmable using C, whereas FireStream still requires assembler language. Intel is also developing a product called Larrabee, but only expects to ship samples at the end of 2008 and make it generally available the following year.