Chip vendor Nvidia plans to use its Cuda parallel computing architecture in all its GPUs (graphics processing units), including its Tegra system-on-a-chip for mobile devices.
Nvidia's Cuda is a C language environment that enables developers to write software to solve complex computational problems by tapping into the many-core parallel processing power of GPUs, according to the company.
The first version of Tegra, scheduled to ship by the middle of next year, will not however have Cuda, said Jen-Hsun Huang, cofounder, president and chief executive officer of Nvidia, in an interview on Wednesday.
Cuda is part of Nvidia's strategy to position its GPUs, traditionally strong in high-end graphics and gaming, as general purpose, parallel computing processors, that can be used in a variety of scientific applications and commercial applications such as financial computing, Huang said.
“We believe that a GPU is not just for graphics anymore, and can be really used for anything that involves a lot of data and mathematics,” Huang added.
Nvidia announced Tuesday a GPU-based Tesla Personal Supercomputer, which it said uses its Tesla GPUs and Cuda to deliver the power of a cluster of computers at a fraction of the cost, in the form factor of a standard desktop workstation. Among the computer makers offering Tesla Personal Supercomputers are Dell, Lenovo, Asus, and Western Scientific.
There is a new computer architecture emerging, and that is based on GPUs and other types of parallel processors, and traditional CPUs (central processing units) working together, Huang said. “The CPU is excellent for sequential processing, but there are many types of problems that you can operate on in parallel”, he added.
GPUs offer higher performance than CPUs as they integrate hundreds of processors, according to Huang. The model of Tesla Personal Supercomputer announced Tuesday, for example, has 240 processors running in parallel, he added.
The first to realize the importance of a “heterogeneous architecture” were gamers who realized that with a CPU and a GPU their video games and 3-D graphics are much better, Huang said.
The GPU in its new positioning is not however seen by Nvidia as an alternative to CPUs. ” We are not trying to replace the CPU as we believe it is necessary,” Huang said.
Nvidia is working with application developers to port their software to the Cuda architecture, Huang said. The ability to program in C language will ensure that sophisticated users like researchers may write the programs themselves for the new supercomputer, he added.