A girl I know wrote gullible on the ceiling of her school. She kept telling people that the word was written on the ceiling - Charlie Demerjian
CHIP DESIGNER AMD has released Firepro S9000 and S7000 GPGPU boards citing single precision floating point performance of four TFLOPS.
AMD's Firepro range of graphics cards are typically found in workstations, but in recent years the firm has been ditching the video outputs and creating GPGPU-only accelerator boards bearing the same branding. The firm announced its latest ultra high-end S9000 and S7000 Firepro accelerators sporting single precision floating point performance of 4.0 TFLOPS and 2.4 TFLOPS, respectively.
AMD's Firepro S9000 and S7000 boards go up against the Nvidia Tesla K10 board that was launched back in May. The Firepro S9000 and S7000 boards sport the firm's latest Graphics Core Next architecture found in its consumer Radeon HD 7970 boards, mated to 6GB of GDDR5 memory.
AMD's Firepro GPGPU boards are a few months behind Nvidia's Tesla K10, both the Firepro S9000 and S7000 and has lower single precision floating point performance. However, AMD's S7000 also has the added advantage of being a single-slot passively cooled board with a TDP of 150W.
Perhaps more concerning for AMD is that Nvidia's Tesla K10 boards come with 8GB GDDR5 memory, a commodity that is extremely valuable in clusters that use accelerators such as the Tesla K10 or the Firepro S9000.
AMD is also working on integrating GPGPUs into its Opteron processors, much like the Firepro branded APU the firm launched earlier this month. Suresh Gopalakrishnan, corporate VP and GM of AMD's server division told The INQUIRER that Opteron APUs will arrive with the next round of Opteron updates.
Even though AMD has penciled in bringing out Opteron APUs, it is unlikely that the firm will be able to stick a 4.0 TFLOP GPGPU into an Opteron processor for a couple of years, meaning that the Firepro S series will still remain the AMD product that provides most of the compute power in a high performance computing cluster. µ
Sign up for INQbot – a weekly roundup of the best from the INQ