logo

EnterTheGrid - Primeur Live!

EnterTheGrid - Primeur is the premier Grid and Supercomputing information source in the world.

>Primeur Magazine
>PrimeurLive!
>EnterTheGrid
>Analysis
>Backissues
>Calendar
>Subscribe
>Advertise
>Contact
News digest 23 June 2005
>Start
>PrimeurLive! from ISC2005 in Heidelberg
>Blog
>Vampir resurfaced
>TOP500
>BlueGene on its way to Jülich Germany
>Hardware
>Bank Sal. Oppenheim explores potential of Linux cluster for financial operations
>The bumpy path to PetaFlop/s scale computing
>Hot Seat Session: transtec to provide hardware, management software and experience
>Hot Seat Session: Cray going strong with X1E, XT3 and XD1
>Trouble in paradise: technology disruptions threaten power and system optimization
>Hot Seat Session: IBM is taking care of the customer, says Mootaz Elnozahy
>Company news
>Quadrics on the Rocks
>UK Atomic Weapons Establishment chooses Linux Networx
Hot Seat Session: Cray going strong with X1E, XT3 and XD1
Heidelberg 23 June 2005

Steve Scott gave an short overview of the three different Cray lines of supercomputers: X1E, XT3 and XD1. He stated that all of them had a measured memory balance with airport cards that deliver high bandwidth. The Cray family of supercomputers consists of very well balanced systems, he said. The company is selling their systems to customers that value these things. Among them are many European customers.

Advertisement
Visit our sponsors
Advertisement

The X1E vector supercomputer installed at the Oak Ridge National Laboratory in Tennessee has already shown great results. The same goes for the system at the US Army HPC research in Minneapolis where it is being used for weather forecasting and CFD. The X1E can also be found at the Korean Meteorological Administration and for

Earth Sciences at the National Institute of Meteorology in Spain. Polar and Marine research is being performed on the X1E at the Alfred Wegener Institute in Germany.

The Cray XD1 is a direct-connect supercomputer. The architecture consists of Opteron AMD processor. The transport connection is normally inserted into a PCI slot but this is not the case with the XD1 where it is feeded directly into the system. This provides a better bandwidth, much stronger communications, and amuch better balance, according to Steve Scott.

The XD1 offers a substantial performance since it is 40% faster than the SGI Altix and 2,3 times faster than the IBM Power4.

The XT3 MPP has a scalable architecture with hypertransport for a direct interconnection, as explained by the speaker. The XT3 is more balanced and powerful than the IBM Blue Gene and the SGI Altix.

The G-HPL Benchmark showed a top performance for the Cray XT3 in four out of five categories. The system had

the fastest recorded result on Linpack and the fastest FFT performance. It also achieved the second fastest global Stream performance. The XT3 offers MPI global random updates.

Early Cray XT3 systems in the United States are Red Storm, the Pittsburgh system, and the one at Oak Ridge. But Cray also sold an XT3 to Japan as well as to the Swiss National Computing Center. It is the first XT3 in Europe, Steve Scott stated.

Inquisitor Resch said that Cray has four special purpose systems: one propriety built system and two Opteron systems: is not that a bit too much? Steve Scott responded that two basic lines of computers is reasonable. Horst Simon asked about the future for vectors and FPGAs. Steve Scott answered that both of them have a good future. There is a new area of architectural innovation. FPGAs have a high potential in conductors but the programming tools are not following as yet. Steve Scott concluded by saying that vectors come en vogue these days.

Advertisement
Advertisement
Leslie Versweyveld

EnterTheGrid - Primeur

James Stewartstraat 248

1325 JN Almere

The Netherlands

http://EnterTheGrid.com

mailto:primeur@hoise.com

© EnterTheGrid - Primeur Live!