Two years ago, both AMD/ATI and NVIDIA realized that GP-GPU (general purpose graphics processing unit) architectures could provide immense processing power, since GPUs are way ahead of CPUs when it comes to crunching numbers and huge amounts of data. NVIDIA came with CUDA last year, and AMD announced that it would release its own GP-GPU architecture codenamed FireStream, although AMD also intends to fuse GPUs with CPUs. AMD launched the FireStream 9170 without making much of a fuss, but it looks like they are ready to launch the second generation, as they have made an official announcement at the International Supercomputing Conference.

Known as the FireStream 9250, the GP-GPU card is specifically designed to accelerate critical algorithms in high-performance computing (HPC), mainstream and consumer applications. The press release informs that the FireStream 9250 easily breaks the one teraflop barrier for single precision performance. It occupies a single PCI slot, for unmatched density and with power consumption of less than 150 watts. This way, AMD’s GP-GPU solution reaches one of the highest performance/watt efficiency with up to eight gigaflops per watt.

The FireStream architecture may be used to calculate financial analyses or seismic predictions faster than with CPU alone. Preliminary tests have shown that there could be 55x performance increases on financial analysis codes as compared to processing on the CPU alone, which supports the efforts to make better and faster decisions.

The AMD FireStream 9250 stream processor includes a second-generation double-precision floating point hardware (the first came with the 9170 model) and its compact size makes it ideal for small 1U servers as well as most desktop systems, workstations, and larger servers. The FireStream-powered cards also feature 1GB of GDDR3 memory, enabling developers to handle large, complex problems.

FireStream 9250, as well as a new SDK are expected to be available in Q3 2008 at an MSRP of $999 USD.