backtop


Print 15 comment(s) - last by peternelson.. on Nov 7 at 8:34 PM

New supercomputer platform

IBM today announced its new System Cluster 1350. The new System Cluster 1350 is fully scalable up to 1,024 nodes. IBM is marketing the System Cluster 1350 as the ideal choice for building a supercomputer. Target markets of the System Cluster 1350 include financial services, industrial, petroleum, life sciences and research.

IBM touts the System Cluster 1350 as: “The IBM System Cluster 1350 provides clients with an arsenal of technologies to tackle the challenges of high performance computing environments, while reducing the complexity and risk of configuring, deploying and managing Linux clusters,” said Wendy McGee, director, IBM Cluster Solutions. “We’ve collaborated with leading-edge processor, accelerator and networking companies to provide clients unprecedented levels of speed, choice and flexibility with this offering.”

As the System Cluster 1350 includes clustering technologies, it’s fully compatible with various IBM products. On the server side of things, IBM multi-core rack mount server’s model x3550, x3650, x3455, x3655, x3755 Intel and AMD based servers. Power 5 servers models p5-505/505Q, p5-510/510Q, p5-550/550Q are compatible too. In addition to traditional rack mount servers the System Cluster 1350 supports IBM Blade Servers models HS21, JS21, LS21, LS41 and QS20 as well.

Since clusters require a high-speed interconnect the IBM System Cluster 1350 supports the latest networking technologies including 10 Gigabit, Gigabit and 10/100 Ethernet protocols. Infiniband 1x and 4x protocols are also supported for up to 10-20 Gigabit data transfer rates. Myrinet is also supported with the System Cluster 1350.

Additionally the System Cluster 1350 is also the first cluster to support QLogic’s HTx adapter powered by PathScale’s InfiniPath HyperMessaging architecture for greater performance improvements. ClearSpeed Advance PCI-X based accelerators are also available in System Cluster 1350 compatible servers. Multiple ClearSpeed Advance PCI-X cards can be installed for linear performance improvements.

IBM’s first customer for its System Cluster 1350 is the Indiana University. Once home to the world’s 23rd fastest supercomputer named Big Red, the Indiana University has replaced Big Red with a System Cluster 1350 cluster. Computing power of the new System Cluster 1350 based cluster is 20.4 trillion numerical operations a second—up from Big Red’s 20 billion.

The IBM System Cluster 1350 is available immediately. Pricing highly varies depending on the configuration.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

First useful comment
By Crazyeyeskillah on 11/7/2006 1:34:50 PM , Rating: 2
Computers like this are bought by the government and defense department. They replace things like the need to test nuclear weapons by creating variables accurate to life that can be tested and built without ever having to detonate a questionable weapon at all. Think about the people that have the most money to spend, and the most to gain from having a technical upper hand. In the medical community, powerful machines like this are giving researchers years of advancements in minutes, by analyzing genetic configurations, variations, and oddities. Products like this can replace "folding at home" type systems to increase security of information, and to give more control to the actual researchers in a more timely fashion.

I just wish we could see things like the secret 'aurora' plane that our government has, think UFO's biotch!




RE: First useful comment
By peternelson on 11/7/2006 8:08:37 PM , Rating: 2
Might also be useful at solving the Valenzetti Equation. Hanso Foundation should fund several for this purpose ;-) They likely do not appear in the top500 list because of the secretive nature of the Dharma Initiative project.

Cluster-based supercomputers are great for many computational applications.

However because of the need for good low latency inter-processor comms, the price per processor or node is relatively high compared to a lot of PCs.

For trivially parallelisable projects which do not require lots of interprocess comms (eg LL primality testing like prime95) it is MUCH more cost effective to just use a pile of COTS (commodity) PCs where price per processor is low.

So depending on your type of application, a supercomputer could be either good value or a waste of money.

I'd likely be interested in IBM's low end cluster systems, but the lack of precise pricing is always a turn-off for me.

Also unless they already use EE (amd energy efficient) or Quadcore (Intel Kentsfield and co) it is likely worth waiting to get better price/performance and running cost effectiveness. And naturally all such processors should be and run in 64 bit mode.


RE: First useful comment
By peternelson on 11/7/2006 8:10:19 PM , Rating: 2
By the way, the article mentions Clearspeed PCI-X boards, whereas Clearspeed will shortly launch a PCI-EXPRESS version of their boards which will offer better performance.

If intending to go with the acceleration tech I'd prefer PCI-E.


"If a man really wants to make a million dollars, the best way would be to start his own religion." -- Scientology founder L. Ron. Hubbard














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki