Posted on 06/27/2007 8:09:14 AM PDT by ShadowAce
Call it "supercomputing smackdown." Today, at the International Supercomputing Conference 2007 (ISC07) in Dresden, Germany, the Top500 group announced its latest list of the top 500 most powerful supercomputers in the world.
To say it's a competitive list would be an understatement and few of the players can hold top spots long without constant improvements. With a few exceptions, this June's list has seen a reshuffling since the last list was published last November.
If there is such a thing as a winner in such a competition, however, that would be IBM (Quote). Just as last November – the top 500 list is recompiled every six months – Big Blue held the top spot, with its Blue Gene/L - eServer Blue Gene Solution. In fact, IBM has taken that top spot four times running with the system deployed by the U.S. Department of Energy (DOE).
"The Blue Gene/L System development by IBM and DOE’s National Nuclear Security Administration (NNSA) and installed at DOE’s Lawrence Livermore National Laboratory in Livermore, Calif., claimed the No. 1 spot," the Top500 group announced Wednesday.
That system achieved a blazing benchmark performance of 280.6 TFlops ("teraflops" or trillions of floating point operations per second) running what's called the Linpack benchmark. (For those who are not mathematically challenged, the Top500 group says the point of the Linpack benchmark is "to solve a dense system of linear equations.")
Interestingly, IBM's performance with the Power-based Blue Gene/L System was identical to its score in November 2006 and in November 2005. (IBM says the system's theoretical performance is 360 TFlops.)
Additionally, however, three other eServer Blue Gene systems made the top ten – last November, IBM only had two of the systems on the list. A fifth IBM system, a BladeCenter JS21 cluster system located at the Barcelona Supercomputing Center in Spain, and also Power-based, came in at number 9, down from the number 5 spot last November.
If that weren't enough, a sixth IBM system, an eServer pSeries p5 575, also made the top ten.
Despite such a strong showing, however, Big Blue doesn't completely dominate. Neither is it the only company whose supercomputers turned in teraflop performances.
Notably, Cray (Quote) supercomputers, powered by AMD processors, hold the number two and three spots on the new list. Both beat the 100 TFlop barrier with more than 101 TFlops apiece. Cray held the number two and ten spots last time.
Intel (Quote) chips are also represented in the top ten supercomputing sites with systems from Dell (Quote) in eighth place and SGI (Quote) at ten. Bumped from June 2007's top ten were Bull SA and NEC/Sun.
Begun in 1993, this is the 29th Top500 list to be published. A lot has changed since then. As a matter of fact much has changed in just the last six months.
"The performance needed to make it onto the list increased to 4.005 TFlops, compared to 2.737 TFlop/s six months ago, [and] the system ranked [at] 500 on the current list would have held position number 216 only six months ago .... the largest turnover rate between lists in the Top500 project’s 15-year history," the Top500 group's statement said.
Trend-wise, despite having only two in the top ten, Intel-based systems continued to gain ground as the predominant high performance computing (HPC) processor, with nearly 58 percent share – 289 out of 500. That's up from 52.5 percent last time.
AMD (Quote) came in second with 21 percent, a decline from 22.6 percent in November, while IBM Power-based systems represent 17 percent, or 85 systems, down from 18.6 percent six months ago. Dual core processors, clusters, and Gigabit Ethernet all dominate, although InfiniBand usage is growing for system interconnections.
The U.S. is home to 281 of the 500 systems on the latest list, and European sites have grown to 127 systems (a jump from 95), while Asian sites dropped to 72 (a decline from 79). Of those, Japan has 23 and China has 13 of the Top500 systems. The UK currently has the most in Europe -- 43 systems -- and Germany is second with 24.
One significant indicator of how dynamic the HPC/supercomputing market is: "The average age of a system in the Top500 list is only 1 year and 2 months," the group said. Also worthy of note, HP (Quote) systems did not rank in the top 50, while IBM systems comprise 46 percent of the top 50.
On the Windows side you priced the cluster pack, which is clustering software that is installed on an existing, licensed copy of Windows. If anyone listened to you in building a new Windows cluster, they'd have paid as little as $700 for Windows Server 2003 Standard * 1,024 = $716,800. Cluster pack is $150 * 1,024 extra, $153,600 for a total of $870,400. The cluster server (W2K3 only licensed to do clustering, clustering software loaded) is about $310 at CDW, so $317,440. You just cost your client $552,960 by not knowing what the hell you're talking about. I really, really hope you have nothing to do with software purchases where you work.
As far as Red Hat, you should, of course, buy Red Hat's HPC cluster license instead of the full server. That's $79 a year, a bit over $200 for three years (the normal Windows upgrade cycle, too, so add Software Assurance to your Windows to equal), about $205,000. And that's without going for a discount like you did with Windows.
So you didn't just cost your client over $100,000 by going with Windows instead of Red Hat for a regular cluster setup, your purchasing practices would have actually cost over half a million. And that with zero hint I know of that Windows actually does clustering faster than RHEL, and REHL is already known to be very, very fast against the regular competition.
In essence, you better have a very specific, targeted reason for going Windows, otherwise you're wasting a lot of money. And Microsoft's argument for "there's more software available on our platform" seriously works against Microsoft in this industry where *NIX rules.
BTW, the last cluster version for Red Hat was a workstation with HPC software, so "That’s a freaking DESKTOP dude.", actually, yes, until the latest version. N3WBI3 was closer than you, and the price was within a few dollars.
Make it four which is more realistic, making it about what the new Windows for clusters costs up front. But of course you ommited all this information about Red Hat costs until now, making your initial whines about Microsoft costs laughable, just like your constant defenses of green party leftists, foreign criminal hackers, and free technology for communist governments.
Microsoft's inability to get Vista out the door in time is the fluke. Not counting that, the NT major release cycle has been on average less than three years going on both the server (1, 2, 4 and 3 years) and workstation (1, 2, 4 and 1 year) branches. And since the new server is expected next year, you'd better get your Windows with SA, otherwise you'll be shelling out for an upgrade in the next year or so.
But of course you ommited all this information about Red Hat costs until now
You were supposed to know, oh expert who immediately started quoting prices. The wrong ones, of course, and said I couldn't be trusted because our prices didn't match (you had the wrong product, and I was going straight from Microsoft's web site). You still haven't apologized for that unwarranted attack.
making your initial whines about Microsoft costs laughable
Windows costs more than Red Hat, and there is no evidence it gives you better performance or management. Why do it unless you just have a hard-on for Microsoft, like you do?
Has he posted where he is getting exchange info on redflag yet?
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.