|

Packing more HPC cycles in to less space with IBM Computing on Demand
IBM’s System x iDataPlex solution is giving organizations a new, more efficient way to scale out supercomputer performance
Email the article Download pdf
In the past, expanding high-performance computing (HPC) capacity almost always meant a larger footprint and a big increase in power and cooling requirements. Often, the space required for expansion was simply not available in the existing data center. IBM’s System x® iDataPlex™ solution is changing the equation for the HPC community, allowing organizations to more easily and economically add HPC computing cycles.
The more energy-efficient and cool-running the solution, the more compute power can be packed into a given space—and the iDataPlex solution makes the most of that fact with energy-sipping features and an innovative highdensity rack. The iDataPlex solution is enabling a wide range of organizations from NASA to the University of Toronto to scale out performance within their data center facilities.
Reducing the HPC footprint by 21:1 with IBM’s iDataPlex With iDataPlex, IBM took the performance and energy-efficiency advances in the Intel® Xeon® processor 5500 series and maximized those advances in a new server design. An iDataPlex rack can hold two times the servers compared to a typical rack of 1U servers. That density comes from the iDataPlex rack design, which provides 84 HPC nodes, plus switches and power distribution units, in a standard two-tile rack footprint. The compute capability that took multiple racks to deploy three years ago can now be deployed using just one iDataPlex rack, enabling organizations to consolidate the HPC footprint by 21:1 or more and reduce power and cooling expenses by greater than 95 percent.
The iDataPlex solution uses less energy to power systems and to cool systems thanks to built-in server, power supply and fan design efficiencies. The iDataPlex rack is also half as deep as a traditional rack, so it takes less power to push air through the unit for cooling. An optional rear-door heat exchanger can even eliminate the need for computerroom air conditioners.
Organizations can leverage a range of networking options with iDataPlex to boost performance as well. The high-speed memory access architecture in the Intel Xeon processor 5500 series helps eliminate bottlenecks, allowing the HPC cluster to even more effectively use networking capability. As a result, many organizations are moving up to 10 Gigabit Ethernet (10GbE) or quad data rate (QDR) InfiniBand to take full advantage of the processing power in their clusters.
Enabling SciNet to run with minimal air conditioning At the University of Toronto’s SciNet supercomputer data center, iDataPlex enables massive computing power combined with exceptional energy efficiency, helping the center earn an impressive 1.16 power usage effectiveness (PUE) rating. That compares favorably to the PUE of 2.5 that the Uptime Institute calculates is average for a typical data center. The SciNet deployment uses the optional rear-door heat exchanger with its iDataPlex systems, which allows the university to run with minimal air conditioning. SciNet expects to dramatically cut power costs as a result. Just as important, the supercomputer will provide capabilities vital to cutting-edge research. With the new supercomputer, says Richard Peltier, scientific director of SciNet and director of the Centre for Global Change Science, “We’re able to do things now that we couldn’t even imagine with the previous systems that we’ve had access to.”
Tripling performance for NASA scientists The NASA Center for Computational Sciences (NCCS) expanded its “Discover” supercomputer by integrating a 4,096-core iDataPlex cluster for a combined performance of 67 teraflops. The new cluster will play a critical role as scientists simulate climate, weather and solar activity.
“By nearly tripling Discover’s performance, NASA scientists will be able to run models with higher resolution and greater fidelity to the underlying physical phenomena,” says Dr. Phil Webster, NCCS project manager and chief of the Computational and Information Sciences and Technology Office at Goddard Space Flight Center. “IBM’s iDataPlex solution for NCCS will provide critical compute power for current and future NASA Earth and space science studies.”
As demand for supercomputer capacity continues to grow, more organizations face the challenge of adding high-performance capacity within tight space and power constraints. IBM iDataPlex gives organizations a new tool to meet that challenge, allowing them to add compute cycles while efficiently managing the HPC footprint and operating costs.
Additional information is available at ibm.com/systems/x/hardware/idataplex
|
 |
|