AMD has broken the one-teraflop barrier for single-precision performance with a graphics card that occupies a single PCI slot. The FireStream 9250 is designed to accelerate critical algorithms in high-performance computing, scientific, and engineering applications -- boosting performance, AMD says, by up to 55 times compared to processing on the system CPU alone.

Scheduled to ship in the third quarter of the year for $999, the FireStream 9250 stream processor tops its model 9170 processor with second-generation double-precision floating-point hardware that delivers more than 200 gigaflops, while consuming less than 150 watts of power. AMD brags that the latter represents a new record in performance-per-watt efficiency (up to eight gigaflops/watt).

For even more formidable HPC development, AMD's archrival Nvidia has unveiled the Tesla 10 Series of GPU programming platforms. The Tesla C1060 ($1,699) is a dual-slot PCI Express 2.0 board that turns a workstation into a 1-teraflop HPC machine, while the Tesla S1070 ($7,999) is a 1U system with four of the new chips, reaching 4 teraflops of performance.