Print 32 comment(s) - last by Lazarus Dark.. on Nov 16 at 3:22 PM

Now imagine if you had to overclock each and every one of those

Supercomputer company Cray announced the availability of its first product in Cray's Rainier program, the Cray XT4, previously code-named "Hood," equipped with AMD Opteron dual-core processors.

According to the press release, the XT4 designed to easily and efficiently scale to a peak performance of more than one petaflops (1,000 trillion floating-point operations per second). Evolved from the Cray XT3 supercomputer, the Cray XT infrastructure provides a common, scalable environment for login, compilation, resource management, work scheduling and I/O. This environment also includes a unique globally shared, high-performance parallel file system, as well as network interfaces to other systems.

"While the theoretical peak speed of supercomputers may be good for bragging rights, it is not an accurate indicator of how the machine will perform when running actual research codes -- which is what our 2,500 users are most interested in," said Horst Simon, director of the NERSC Division at Berkeley Lab. "To better gauge how well a system will meet the needs of our users, we developed SSP, a sustained system performance benchmark suite. Under this real-world performance test, the new Cray XT4 system will deliver over 16 teraflops on a sustained basis."

The Cray XT4 supercomputer uses up to 30,000 AMD Opteron dual-core processors interfaced to the Cray SeaStar2 interconnect chip. Unlike typical cluster architectures, in which many microprocessors share one communications interface, each AMD Opteron processor in the Cray XT4 system is coupled with its own interconnect chip. Providing six links in three dimensions, the SeaStar2 chip uses its embedded routing capability to take advantage of HyperTransport technology and accelerate communications among the processors

"Unlike commodity clusters, the Cray XT4 supercomputer is built from the ground up to provide a scalable, balanced system in order to support the most demanding applications," said Cray President and CEO Peter Ungaro. "The Cray XT4 system builds upon our highly successful Cray XT3 supercomputer in almost every aspect, providing enhanced scalability, performance and reliability in a system that is easily upgradeable to protect a customer's investment for years to come."

Cray XT4 supercomputers configured with dual-core Opterons are available now. Furthermore, Cray says that current XT4 systems will be upgradable to AMD's quad-core processing technology when available.

Comments     Threshold

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

By Orpheus333 on 11/15/2006 12:21:34 PM , Rating: 5
The reason for AMD CPU shortages? hahaha

RE: hmmm
By vbNetGuy on 11/15/2006 12:39:58 PM , Rating: 2
Thought the exact same thing!

RE: hmmm
By archcommus on 11/15/2006 1:16:20 PM , Rating: 2

Cray: Can you give us about 30k Opterons? kthx
AMD: Uh...

RE: hmmm
By gooser on 11/15/2006 4:15:43 PM , Rating: 2
coming right up!

By ixelion on 11/15/2006 12:03:50 PM , Rating: 5
Will it run Crysis at max settings?

RE: nope
By MADAOO7 on 11/15/2006 12:43:00 PM , Rating: 3
As long as you turn off anti-aliasing.....haha

By encryptkeeper on 11/15/2006 1:20:17 PM , Rating: 2
Will it be Vista capable?

RE: Leaks?
By zinfamous on 11/15/2006 1:29:29 PM , Rating: 2
Without a decent GPU? Looks like it won't be!

RE: Leaks?
By retrospooty on 11/15/2006 10:39:23 PM , Rating: 3

Old news
By 9nails on 11/16/2006 12:34:06 AM , Rating: 2
Big deal! Sony said that they can do numbers like that with one PlayStation 3.

RE: Old news
By Alpha4 on 11/16/2006 1:25:06 AM , Rating: 2
Heh! I think Sony suggested the PS3 could calculate FPmath at a theoretical 2 teraflops, so technically a beowulf hub of 500 PS3s = 1 Cray XT4 or 30,000 Opterons oO. I'm not inclined to believe thats perfectly true.

RE: Old news
By paydirt on 11/16/2006 10:21:58 AM , Rating: 2
Folding@Home is generating 35 Teraflops from 532 CPUs which means there are 532 users that are crunching numbers by using at most 1064 ATI graphics cards. 1064 is assuming that each person with a registered CPU is crunching numbers using ATI cards in a SLI configuration. (The heavy lifting--number crunching is done on the GPUs)

...but the Cray does those teraflops non-parallel, so it has its uses.

By priestx on 11/15/2006 5:11:02 PM , Rating: 2
Yet it STILL can't divide by 0.

By MrDiSante on 11/15/2006 6:11:15 PM , Rating: 2
It'll give you Double.NaN or Double.infinity.

Time to sue Cray.
By SilthDraeth on 11/15/2006 5:12:43 PM , Rating: 2
My name is Rainier, and I did not give them permission to name a program after me.

RE: Time to sue Cray.
By KristopherKubicki on 11/16/2006 3:14:31 AM , Rating: 2

Please cease and desist in any useage of the word "Rainier" as we have significant prior corporate usage of the term.

Washington, USA

By AxemanFU on 11/15/2006 4:55:13 PM , Rating: 2
Processors with dozens or hundreds of pipelines, adapted from the GPUs of today and tomorrow like the latest Radeons and Geforces will probably take over a whole lot of the supercomp market from general purpose CPU processors because they can work so much faster doing massively parallel mathematical calculations. For example, 1900 series Radeon cards running the latest version of folding@home blows the doors off of any current CPU, even the quad core intel procs.

Once they can work out the interface issues to allow these to be used in large arrays instead of only one or two at a time, there should be a big shift to these more specialized FPmath focused processors.

only dual?
By pepsimax2k on 11/15/2006 5:04:19 PM , Rating: 2
what, no quad core? damn, obselete already. anyone need 30,000 processors?

I wonder...
By Spartan Niner on 11/15/2006 5:23:04 PM , Rating: 2
How many ppd would this give me in F@H? ;)

By Harkonnen on 11/15/2006 5:36:44 PM , Rating: 2
"the Cray XT4, previously code-named "Hood," equipped with AMD Opteron dual-core processors."

Where is the Bismark?


But the real question is...
By judasmachine on 11/15/06, Rating: 0
RE: But the real question is...
By quickk on 11/15/06, Rating: 0
RE: But the real question is...
By SilthDraeth on 11/15/2006 5:24:34 PM , Rating: 1
At least do it right.

In Soviet Russia the question answers you.

in other news..
By lwright84 on 11/15/06, Rating: -1
RE: in other news..
By saratoga on 11/15/2006 12:41:59 PM , Rating: 5
Opterons are used because of the HT links. You're not going to get this kind of performance outside of Core2s because systems like this are generally limited by interconnect more then core performance, and the Core2 doesn't stand much of a chance here. Thats why you see lots of supercomputers based off old IBM PPC cores that are outclassed by low end desktop chips.

Where Core2 would make sense is in clusters.

RE: in other news..
By finalfan on 11/15/06, Rating: 0
RE: in other news..
By Lonearchon on 11/15/2006 3:47:10 PM , Rating: 3
According to the diagram that is attached to the article the Cray interconnect chip SeaStar2 connects to the system via a HT link allowing 6.4gbs from the system to the interconnecting network

RE: in other news..
By MonkeyPaw on 11/15/2006 5:43:34 PM , Rating: 2
I think Opteron is popular in such machines because it has both an IMC and HT. HT is nice because it's a free and open high-speed bus. Use of HT is pretty much required, even if it is just used to tap into an dedicated hub, as HT is how the K8 communicates with the rest of the system (and why all K8 platforms perform remarkably similar). Really, I think more credit in this case should go to the IMC, as it allows each CPU to have access to its own low-latency memory. The other SMP solutions require some sort of shared bus, which becomes increasingly inefficient as you add more CPUs to the system. I think the lack of an IMC is why you don't see Core-based supercomputers at this point. Intel is supposed to resolve this in time, as the next server chipset is supposed have an HT-like bus that allows for fast communication between shared memory controllers. Eventually, Xeon will get its own IMC, and that should really remove the server barriers for Core, but that's not supposed to be introduced until at least 2008.

RE: in other news..
By Jkm3141 on 11/15/2006 11:27:05 PM , Rating: 2
yea, the HT and IMC allows the opteron to be deployed easier. the IMC takes care of having a separate memory controller for the processors, and the HT allows an open and easy to implement connection between the CPU and the proprietary connection between each of the 40,000 cpu's or whatever it is. HT is not that scalable, but much easier to work with than Intel's closed FSB to connect to a more scalable interconnect.

RE: in other news..
By Lazarus Dark on 11/15/06, Rating: 0
RE: in other news..
By Spartan Niner on 11/15/2006 5:21:43 PM , Rating: 3
This is not a cluster system! A supercomputer is based upon intercommunicating, connected processors which form the equivalent of one mega-processor. A cluster is based on the concept of parallelism and splits up work into chunks for calculation by each node of the cluster. Therefore, supercomputer != cluster.

RE: in other news..
By Lazarus Dark on 11/16/2006 3:22:26 PM , Rating: 2
your right. must have been tired. please replace cluster with supercomputer and the my statement still stands.

"Paying an extra $500 for a computer in this environment -- same piece of hardware -- paying $500 more to get a logo on it? I think that's a more challenging proposition for the average person than it used to be." -- Steve Ballmer
Related Articles
AMD Q4'06 Dual-Core Roadmap
October 3, 2006, 8:23 AM
AMD Announces Quad-Core Tape Out
August 15, 2006, 2:47 AM

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki