Spanning Physical Boundaries of GPU Computing with Kepler

Spanning Physical Boundaries of GPU Computing with Kepler

Kepler Virtualized and Ready for Cloud Computing

Leveraging on the buzz and excitement of his opening keynote address at GTC 2012, NVIDIA CEO Jen-Hsun Huang revealed the full capabilities of the Kepler GPU. He also spoke in length about NVIDIA's plan to put Kepler in the clouds, by enabling the parallel computing prowess of the Kepler GPU to be distributed over cloud computing services.

NVIDIA CEO Huang was practically walking on cloud nine as he spoke about the cloud computing capabilities of the Kepler GPU during his keynote address at GTC 2012.

Although we have briefly touched on the features of Kepler during the launch of NVIDIA GeForce GTX 680; however, it did dawn upon us that Kepler was meant for bigger things. CEO Huang did reveal more features of the Kepler architecture by positioning it as the world's first virtualized GPU.

With its Virtualized GPU technology, the physical command buffers have been virtualised. Each graphics command that the buffer receives goes through a MMU (memory management unit) lookup and does a virtual physical-to-virtual address translation. This enables the GPU to ascertain which virtual machine has sent the graphics command. The shared Kepler GPU is able to render the command into the frame buffer of that specific virtual machine and send it back accordingly.

By breaking down the traditional dedicated relationship of a GPU and a user with its Virtualized GPU technology, NVIDIA has readied Kepler GPU to be deployed to cloud computing and its related services.