Loading ...Nestled just north of the Italian border, in southern Switzerland, you will find CSCS, the Swiss National Supercomputing Center. Long known for chocolate and breathtaking Alpine landscape, Switzerland is also known in the supercomputing circles as an HPC powerhouse.
We traveled to Manno, just north of Lugano to meet with some of the brightest minds in HPC, as well as unveil their latest supercomputer, Monte Rosa – named after the Swiss mountain, the tallest in the regional border with their Italian neighbors.
Monte Rosa is based on a Cray XT5 platform, which is quite popular with the supercomputing crowd because of its massively scalable architecture and high-throughput interconnects. Monte Rosa features 14,762 processors, capable of delivering up to 141 teraflops of peak performance.
To complement the huge number of processors, 29.5 terabytes of main system memory are available for computation. Of course you need somewhere to store all of the results, so a 290 terabyte storage system holds the results from processing runs.
With reported performance of nearly 10 times that of its predecessor, the new Monte Rosa is liquid cooled, allowing it to fit in the same physical space, helping to optimize the center’s floorspace.
The system was installed in record time in May of this year, thanks to Cray’s modular engineering efforts, allowing the center to begin immediately reaping the rewards of the system. Within only a few days of bringing the system online, it was already being utilized near its full capacity.
The productivity seen with the new system is expected to help a variety of industries within Switzerland. While this system is based on Quad-Core AMD OpteronTM processors today, there is already a planned upgrade to Six-Core AMD OpteronTM processors before the end of the year, bringing the total performance to over 200 teraflops.
While some computing centers focus on building out capacity, CSCS likes to focus on the applications. They’ve built up an impressive staff of technologists who can not only focus on optimizing the supercomputer platform, but also, spend most of their time in the applications, where they believe they can deliver their true value-add. Science and engineering applications such as climate, weather, biology chemistry, physics and material sciences are all aided by this new cluster.
But that is not to say that they haven’t built out some amazing capacity. Monte Rosa is now the 23rd largest supercomputer in the world and the 4th largest in Europe according to the June 2009 Top 500 list.
With the performance that they are seeing using Quad-Core AMD Opteron processors, we can’t wait to see what happens when they are able to increase capacity with the Six-Core AMD Opteron processors.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
(4.93 out of 5)
#1 by Datsun - September 24th, 2009 at 13:21
How does it perform in double precision floating point? How many significant performance penalty in percentages compared with single precision?
#2 by John Fruehe - September 24th, 2009 at 15:27
I don’t believe I have that data.
#3 by Joshua Mora - October 13th, 2009 at 23:59
quad core and six core issues 4 flops/clk/core
multithreaded and single threaded DGEMM and SGEMM efficiencies are in the 90%. HPL (~DGEMM) in the 85% within node, 80% in cluster.
Based on comments from John, 141TF is peak at 2.4GHz Shanghai. So real TFLOP double precission should be in the 80%, ie. 113TF.
When upgraded to six core, say at 2.6GHz (eg. at same power consumption per processor), you should be able to get 6/4 * 2.6/2.4 increase , ie. 1.625 (62% increase), which is 221 TF peak, and 211 TF peak if using 2.4GHz, with similar efficiencies ~170TF real at 2.4GHz six core.
Not bad for an upgrade of processors.
#4 by james - October 3rd, 2009 at 06:09
With reported performance of nearly 10 times that of its predecessor, the new Monte Rosa is liquid cooled, allowing it to fit in the same physical space, helping to optimize the center’s floorspace.
————–
is the liquid cooling system similar to(although much larger in scope) what’s been on desktops or is it an entirely customized cooling solution?
thanks!
#5 by John Fruehe - October 4th, 2009 at 19:56
No, the liquid cooling that is used happens outside of the chassis. It is a raised floor data center. Cool air comes from under the floor, is forced up through the rack to cool the systems. When it gets to the top it chilled with cold water. The system is designed around removing the ambient heat inside the system.
Liquid cooling in desktops happens inside of the chassis and is usually directly connected to the processor.