Hewlett-Packard and Dell are continuing to build servers that bring graphics chips into the data center.
HP officials on May 17 announced that Nvidia's new Tesla M2090 GPU (graphics processing unit) will appear in a number of upcoming HP ProLiant servers aimed at the HPC (high-performance computing) space, a move they say will address organizations' demands for greater performance and energy efficiency.
The announcement from HP and Nvidia came a day after officials from Dell and Advanced Micro Devices, at the Microsoft TechEd show, said the server maker will roll out the PowerEdge M610x blade, which will include AMD's FirePro V7800P GPUs. Officials with both Dell and AMD said the combination of Dell blades with AMD GPUs will help in such areas as HPC and desktop virtualization.
HPC organizations have been particularly strong adopters of GPUs for computing tasks. For workloads optimized for parallel processing, companies can get much more performance with GPUs-which offer hundreds of cores-than with traditional CPUs alone, and without increasing power consumption.
Further reading
"The high performance computing (HPC) segment has an endless thirst for performance and this has made the use of GPUs a disruptive force," Ed Turkel, marketing manager for HPC for HP, said in a May 17 blog post on the Nvidia Website. "For researchers, scientists and engineers, enhanced performance allows for faster innovation that will result in the kinds of discoveries that can change the world."
Nvidia officials called the Tesla M2090 the world's fastest parallel processing chip, which can hold as many as 512 cores and can offer application acceleration 10 times that of CPUs alone. The Tesla M2090 GPU will be available on HP's new ProLiant SL390 G7 4U server, which Nvidia officials said was built specifically for hybrid computing environments that use both GPUs and CPUs.
According to HP's Turkel, the Tsubame 2.0 supercomputer in Japan-named the world's fourth-fastest supercomputer on the Top500 list in November 2010-is an example of what GPU computing can offer. The system uses 1,408 ProLiant SL390s G7 servers, and delivers similar peak performance as Cray's XT5 Jaguar system at the Oak Ridge National Laboratory. However, Tsubame 2.0 does all that with 92 percent fewer servers and consumes one-seventeenth of the power than Jaguar, which is powered by AMD's Opteron chips.
"Take that into consideration and it's easy to see why GPUs have made such an impact in HPC," Turkel said.
AMD officials said Dell's new server is the FirePro's first foray into the data center. The FirePro V7800P is designed to offer fast performance and massive parallel processing in what officials said is a passively cooled form-factor. Like FireStream GPUs launched a year ago, the FirePro V7800P are designed without fans; instead, the heat sinks use air flow in the server chassis for cooling.
With the new offering, Dell and AMD are continuing what has been a successful partnership over the past few years, according to Brian Payne, executive director of platform marketing for Dell's PowerEdge server portfolio.
"We're collaborating together to drive innovation around graphics processing solutions and you won't find a better example than our Dell PowerEdge M610x blade," Payne said in a statement. "This blade helps support and accelerates multiple remote users while providing a rich graphical experience and maximum hardware asset utilization."
In addition to graphics, AMD's FirePro V7800P can be used for a wide range of workloads, according to Sandeep Gupte, director of professional graphics product management at AMD.
"The AMD FirePro V7800P is AMD's first multifunctional server solution for compute, VDI [virtual desktop infrastructure] and professional graphics offering, IT departments GPU compute for high-performance computing matched with leading-edge graphics virtualization," Gupte said in a statement. "The AMD FirePro V7800P allows AMD to draw upon a powerful professional graphics solution to meet the needs of an entirely new set of data center customers."
Research Assistant
