|
NVIDIA demonstrated a computational version of Kepler GPU , which has become the accelerator called Tesla K10. The decision is traditionally designed for high-performance computer systems and is available for order through NVIDIA channel partners in the face of Appro, Dell, HP, IBM, SGI, and Supermicro.
The accelerator has two on-board graphics processor, but the list of technical specifications suggests that the base for Tesla K10 is a dual GK104. Thus, each GPU has 1536 cores CUDA, coupled with an array of GDDR5 memory chips totaling 8GB with capacity 320 GB / s and support for error correcting codes (ECC). Performance is measured by the value 4.58 teraflops in single precision calculations.
Separately, NVIDIA plans to actively promote a product called VGX, which is a card with four GPU and 16 GB video memory. The solution supports hardware virtualization, and focused primarily on cloud gaming services (such as OnLive, GaiKai or Playcast). Experts NVIDIA develop their own cards for gipervvizor VGX, which can work with the software Citrix and VMWare. Later will be added support for Microsoft products and Xen.
We add that in autumn this year nvidia will release Tesla K20 accelerator , which is optimized for double precision calculations. Related Products :
|