AMD FireStream 9170

AMD has announced the FireStream 9170, a graphics processing unit (GPU) aimed at intensive computing. It supports double-precision (64 bits) floating point (FP) numbers, which is required by some researchers just to consider developing on a specific platform. While many applications have been developed with single-precision numbers (as demonstrated by rival NVIDIA), other computing techniques simply require double-precision to work.

As usual, the press release is written in PR jibberish, but in short, it’s a paper-launch (no product available yet) and it is not clear if the 500 GFlops mentioned in AMD’s press release have been theorized using single or double precision numbers.