14 DAY TRIAL // AMD has been reported to change the classical approach towards their GPUs in the forthcoming R700 series. The nowadays GPUs use parallelism for heavy computing, while preserving their monolithic structure and the same number of stream processors, texture units and so on. AMD is determined to give up the current technology in favor of a multi-core architecture.
The R700 cards will all feature GPUs with variable numbers of R700 cores, according to the market segment they are addressed to. For instance, AMD's state-of-the-art graphics card will allegedly have four or more cores included in a single coat of die, with an average computing performance of two trillion floating-point operations per second (two teraflops, that is).
The speed difference between the R700 powered cards will be done by specifying the number of integrated cores. Less cores would mean a low-end graphics card, while a superior core count would label the card as "professional" or high-end. The actual Radeon HD 3870 provides only quarter the estimated speeds for the next generation of video cards, namely 500 gigaflops at most.
Of course, performance doesn't reflect in the flops count, but chances are that the R700 series will be incomparable faster that the nowadays GPUs.
Surprisingly or not, the rumor couldn't have sprung at a more appropriate time, as Intel has just disseminated a similar information: we are talking about the Larrabee project, that is expected to finalize into a game-oriented graphics processor that will make its presence somewhere in the near future.
In early April this year, Intel was mentioning something about the new GPU featuring multiple, small "throughput cores", while images that leaked from within Intel's workshops showed a graphics chip that integrated almost a dozen of cores with a 4-megabyte shared cache between them.