Posted on 11/08/2004 9:34:31 PM PST by m3d1um
U.S.A tops supercomputer list again!! #1 and #2 beating Japans "Earth Simulator". Topping the charts is IBM and the US Department of Energy's 'BlueGene/L DD2' beta system, at 70.72 TFlops, followed by NASA's 'Columbia' at 51.87.TFlops. Go USA!!!!!!
Americans should rue the day
the government spurned Seymour Cray!
A "flash mob supercomputer" event was staged this past summer in San Francisco, attempting to crack the top 500 list. All sorts of folks hauled in PCs of all types and they were linked together. The benchmark was Linpack. They failed to reach the requisite speed because of technical problems, but it was a very good showing nontheless.
Since then, the Cray name and other portions of its expertise were sold, to another company. But many of the original Crayons remain with SGI.
As a yardstick of performance we are using the `best' performance as measured by the LINPACK Benchmark. LINPACK was chosen because it is widely used and performance numbers are available for almost all relevant systems.
The LINPACK Benchmark was introduced by Jack Dongarra. A detailed description as well as a list of performance results on a wide variety of machines is available in postscript form from netlib. Here you can download the latest version of the LINPACK Report: performance.ps. A parallel implementation of the Linpack benchmark and instructions on how to run it can be found at http://www.netlib.org/benchmark/hpl/.
The benchmark used in the LINPACK Benchmark is to solve a dense system of linear equations. For the TOP500, we used that version of the benchmark that allows the user to scale the size of the problem and to optimize the software in order to achieve the best performance for a given machine. This performance does not reflect the overall performance of a given system, as no single number ever can. It does, however, reflect the performance of a dedicated system for solving a dense system of linear equations. Since the problem is very regular, the performance achieved is quite high, and the performance numbers give a good correction of peak performance.
By measuring the actual performance for different problem sizes n, a user can get not only the maximal achieved performance Rmax for the problem size Nmax but also the problem size N1/2 where half of the performance Rmax is achieved. These numbers together with the theoretical peak performance Rpeak are the numbers given in the TOP500. In an attempt to obtain uniformity across all computers in performance reporting, the algorithm used in solving the system of equations in the benchmark procedure must conform to the standard operation count for LU factorization with partial pivoting. In particular, the operation count for the algorithm must be 2/3 n^3 + O(n^2) floating point operations. This excludes the use of a fast matrix multiply algorithm like "Strassan's Method". This is done to provide a comparable set of performance numbers across all computers. If in the future a more realistic metric finds widespread usage, so that numbers for all systems in question are available, we may convert to that performance measure.
Big Blue BUMP!
Mac Supercomputer at Virginia Tech PING!
I see the Virginia Tech Apple Mac supercomputer is now at number 7... down from number 3 last year. It is also faster than it was last year.
Hmmmm... seems like some on here were claiming the VT Apple would be WAY down the list with bunches of Intel/Windows supercomputers beating it out.
The only Intel supercomputer on the list above the Macintosh machine is number 5, the Intel Itanium THUNDER at the Lawrence Livermore National Laboratory... and it took almost twice as many processors to do it. Bet it cost a heck of a lot more than VT's computer.
If you want to be included or excluded on the Mac Ping List, freepmail me.
Not true. The list isn't clear, but the NASA machine is powered by Intel.
The list is quite clear, but he's probably thinking x86 Intel. I guessed 7 or 8 for the VA Tech system and was right, but the surprise here for me is that I expected some Opteron-based Cray XD1s in the top 50. I guess there wasn't enough time between release of the product and this list for someone to make a big enough system.
Other interesting goodies:
The US Army bought an xServe cluster that's faster than the VA Tech system, but it's not on the list. And there are several more Mac systems waiting to be on the next Top 500. Education seems to be snapping these things up.
The Macs give over 2 MFlops per dollar, while the nearest Itanium, Opteron and Xeon competitors run at best half that, usually far less.
Be sure to ping me when they start giving "bang for the buck" trophies at Indianapolis and Daytona... ;)
Race cars just race. These things are actually supposed to do some work, where "bang for the buck" can be important.
I'm guessing that Apple's good price/performance ratio and their recent packaging of XServes into the Apple Workgroup Cluster will have us seeing lots of educational and pharmaceutical XServe clusters on the Top500 in the next year. Probably not too highly placed, but they'll get their spots starting probably at <$500K.
bttt
And more FLOPS = more work. I'm sure NASA will find a use for those extra cycles somehow...
I'm sure they will. But they paid nine times the price for five times the performance. Of course, horses for courses -- maybe that Itanium system is faster for their specific work. But it's nice to see how cheap and easy the XServe has made cluster computing.
Errr, well, if we're to take the LINPACK results at face value - I personally don't consider LINPACK to be a very worthwhile benchmark, but there you go - it's faster for most any sort of work. The question is whether that increase was worth the price - apparently, to NASA, it was.
But it's nice to see how cheap and easy the XServe has made cluster computing.
Agreed. It'll be interesting to see where the upcoming AMD machines land...
AND it looks like they got that five times speed by using five times the processors... it makes you wonder what would happen if you built an XServe Cluster with 10,160 processors.
Certainly cheaper... perhaps faster.
Not likely - the NASA machine is a true supercomputer, not a cluster. The reason the NASA machine costs so much more is not because the I2 is dramatically more expensive than the G5 - the chips are the cheap part of the equation. What differentiates the two, and where SGI earns its money, is in the interconnects that link all those processors together. The Mellanox stuff that VT uses is a nice, off-the-shelf solution - it's reasonably fast and reasonably cheap, but it's not going to scale anywhere near as well as the SGI machine's NUMAflex architecture. At some point, the interconnects will start getting swamped with the IPC overhead needed to coordinate all those processors, and then it doesn't matter how fast the processors are - doesn't do you any good to have the fastest chips in the world if you can't feed data to them.
In order to get performance at that level, with that number of chips, you're out of range of simple, off-the-shelf clustering interconnects, and into the land of exotic, high-speed, proprietary shared-memory interconnects, and that costs big bucks - the NASA machine is a shared-memory machine, which the VT cluster isn't, and can't be, because the interconnects aren't fast enough for it. The bus logic alone on that NASA machine probably cost more by itself than all the Intel processors put together. If you want a G5 supercomputer to match the performance of the NASA machine, it's going to take more than just shoveling a few thousand more chips into the box.
Not quite. Columbia is a cluster of 20 SGI Altix mainframes, each with 512 processors and 1TB shared memory. The nodes are still hooked up together with infiniband, just like System X. Although I'm sure you're right in that each 512 processor node is internally more efficient than 256 individual XServes.
Quick, send the plans to China. We can't be allowed to 'lead' the world.
Gee, where's Xlinton when we need him?
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.