Appro's new HyperPower Cluster, announced Monday, mixes the processing power of CPUs and graphics processing units to tackle high-performance computing tasks.
The server cluster is designed to break up and simultaneously execute thousands of threads and tasks across servers with Intel CPU cores and Nvidia GPU cores.
Graphics cards are considered more advanced in tackling scientific and math applications than CPUs, which are designed to handle generic computing tasks. Many PCs, for example, come with separate CPUs and GPUs in which specific multimedia tasks -- like video editing -- are unloaded directly to graphics chips, leaving CPUs free to execute tasks like word processing and virus scanning.
The main job of CPUs in the cluster will be to analyze code and then off-load complex processing tasks to the GPU, said John Lee, vice president of the advanced technology solutions group at Appro. The cluster includes 304 cores with quad-core Xeon chips and 18,240 Tesla S1070 GPU cores that provide 78 teraflops of performance.
The cluster includes multiple 1U servers integrated in a 42U standard rack configuration. It is targeted mainly at institutions that need high-performance computing capabilities, like academic institutions and oil and gas exploration companies, Lee said.
In current computing environments, GPUs are highly parallel processors designed to run repetitive tasks like graphics processing, said Dean McCarron, principal analyst at Mercury Research. CPUs, on the other hand, are general-purpose in nature and can take the load off GPUs by running basic applications.
"Either one could technically do the job of the other. But they each have areas that they are better processing than others," McCarron said.
Though Intel and AMD maintain that CPUs are essential to PCs, Nvidia has pushed GPUs as an alternative to CPUs. Nvidia says that CPUs are slow, and GPUs execute advanced tasks like video encoding and decoding much quicker.
"The CPU has run its course and the soul of the PC is shifting quickly to the GPU," said Jen-Hsun Huang, Nvidia's CEO, in February when defending the company in a licensing spat with Intel.
In the future GPUs might integrate capabilities of CPUs and vice versa, McCarron said. A number of chip makers are indirectly taking steps to bring graphics capabilities closer to CPUs. Later this year Intel will start shipping Westmere chips that merge the CPU and GPU in a single package. Advanced Micro Devices is scheduled to release chips in 2011 that merge GPUs and CPUs on one chip.
Intel is also developing a chip code-named Larrabee that combines the processing power of many x86 cores to tackle graphics and supercomputing applications.
Pricing for Appro's cluster of servers starts at US$250,000 and goes up to $1 million, Lee said. The product is available worldwide.
This is not the first product to rely on GPUs to provide processing power. One Nvidia GPU-based supercomputer, the Tsubame computer at the Tokyo Institute of Technology in Japan, has more than 30,000 processing cores that provide a processing speed of 77.48 teraflops. It was ranked the 29th-fastest supercomputer in the world in the Top 500 ranking in December.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.