The IBM iDataPlex Dx360 M3 is powered by both Intel Xeon central processing units (CPUs) and Nvidia Tesla graphics processing units (GPUs) and is designed to be clustered with other Dx360 M3 modular servers to form a supercomputer.
This is the first time a major computer company has adopted Nvidia GPUs for a supercomputer that can be marketed worldwide.
“Now customers will have access to everything IBM brings to the table in high-performance computing,” Sumit Gupta, senior manager of Nvidia’s Tesla, said in a phone interview Monday.
In a supercomputer, the CPU and GPU work together in a way that is analogous to a PC, according to Gupta. In the case of a PC, the CPU handles, for example, the launching of applications and getting files from the disk, while the GPU handles highly specialized computational tasks. Likewise, when a supercomputer does certain computation-intensive tasks, like calculating the behavior of many water molecules, a GPU can be more suitable than a CPU, Gupta said.
IBM iDataPlex Dx360 M3 is a supercomputing design powered by both Intel Xeon CPUs and Nvidia Tesla GPUs. The two Tesla GPUs are on the right. Each GPU requires around 200 watts of power and a 16-lane PCI-e connection.
“While the CPU orchestrates the system activities…the GPU is an idiot savant: it computes the mathematical kinds of things,” David Turek, vice president of deep computing at IBM, said in an interview.
And though this is the first time IBM has offered a high-performance computer with an Nvidia GPU, it is not the first hybrid design at IBM. Big Blue’s Roadrunner supercomputer was a hybrid, according to Turek.
“This is not the first hybrid GPU system in the world–they’ve been deployed for some time by other companies,” he said. “But the difference is twofold. One is the passage of time, which has allowed the marketplace to mature in its perspective of GPUs–where the GPU is advantageous, where it’s limited. And (two) the fact that it comes from IBM means that you have a company with some mass behind it to back up customer initiatives,” Turek said.
And though neither IBM nor Nvidia will cite customers at the moment, they are big names, according to Turek. “We do have customers, and they are significant. Wall Street, oil and gas exploration, universities, and government labs,” he said.
Pricing was not disclosed.