NVIDIA and AMD GPUs to announce its new supercomputers
Currently cGPUs higher strength are Nvidia Tesla M2090 (0665 TFlops in double precision calculations) and amd FirePro S9000 (0806 TFlops in double precision calculations) but we know that Nvidia plans to take the crown of that market with its still not released Tesla K20, which offer 1.13 TFlops in double precision calculations .
Even without being released, Tesla K20 cGPU not the more powerful Nvidia will offer, because today announces the new Tesla K20X (also based on the core kepler GK110 ), but unlike Tesla K20, SMX will have 14 active, is say 2688 shader processors (Tesla SMX K20 has 13 shader processors and 2496) and will be equipped with 6GB GDDR-5 memory ECC, this setting allows you to offer up to 1.31 TFlops in double precision calculations.
AMD Nvidia does not intend to let him snatch the performance crown, and also announces its new FirePro cGPU SM10000, which will have a yield of 1.48 TFlops in double precision calculations, and will be equipped with 6GB GDDR5 ECC memory. Although unaware of the specifications of this new product, previous rumors suggest that it could be a dual-GPU solution based on two Tahiti LE AMD GPUs (Graphics Core Next architecture), but nothing is confirmed yet about it.
Both new products are aimed at mission-critical tasks, GPU virtualization, streaming, scientific calculations, among many other tasks of server and supercomputer. Neither Nvidia nor AMD have revealed the release date or price for these new products.
Link: Nvidia, AMD graphics processors for supercomputing release (ComputerWorld)Tags: amd, card, cGPU, FirePro, GCN, Graphics Core Next, K20X, kepler, Nvidia, server, SM10000, supercomputer, Tesla, video