“The Army Research and Development Command will use a giant cluster of Apple Computer Inc.’s G5 servers [Xserves] to build one of the fastest supercomputers in the world to research the aerodynamics of hypersonic flight,” Brian Robinson reports for Federal Computer Week.
“The MACH 5 (Multiple Advanced Computers for Hypersonic research) supercomputer, announced earlier this week, will use 1,566 of the 64-bit dual-processor servers and is expected to top 25 teraflops per second when it comes online later this year. The fastest supercomputer in the world now is Japan’s Earth Simulator with a maximum performance of just less than 36 teraflops,” Robinson reports.
“MACH 5 will cost $5.8 million to construct, a fraction of the price purpose-built supercomputers bring. The Earth Simulator cost around $350 million. Apple won the Army contract after a competition among half a dozen companies based on such things as power requirements, cooling needs and floor space requirements, as well as performance,” Robinson reports.
Full article here.
OSX plus 970PPC: Unbeatable.
Yeah, but what if they get a virus.
Incredible throughput. It amazes me.
Yeah, what Sputnik said.
What virus….there are none.
wow! we got sputnik and one-nut posting on the same article!
hey guys, how’s it going? must be a slow news day today. nothing witty yet to have fun with…
Sputnik you must have the windows head cold
I think Sputnik was joking, thereby showing how much better a solution this is than windows, on top of the speed to $ performance.
How many super cluster gigaflops does it take to model how much force required to beat a terrorist into submission without leaving any tell-tale marks for the Red Cross inspection?
This is really cool. I can’t wait to hear how well it does, but hopefully it won’t be Classified.
“Yeah, but what if they get a virus.”
Then they can call Sophos.
Just buy off the Red Cross. They are always looking for donations.
Good News for Apple & IBM
When the MACH 5 and VT Clusters are online later this year Apple should have 2 of the 10 fastest Supercomputers in the world. Why is it good for IBM? IBM is already a huge player in this area and the G5 processor is an IBM product. The PPC design is making major strides these days and there will be more to come.
No one here caught the obvious error?!
“… is expected to top 25 teraflops per second”
1 flop = 1 instruction per second
1 teraflop = 1 trillion instructions per second
1 teraflop per second = 1 trillion instructions per second per second
I hardly think the author is talking about the rate at which instruction execution is being accelerated, but was rather being ignorantly redundant.
It was correctly used later, “…just less than 36 teraflops”.
Way to go Apple! Go Army!
Here’s an idea. They should get the Apple supercomputer to finish developing Longhorn for Windows and charge em that hefty 56 billion bank account they got.
BTW, where’s my AI software? There is now the hardware out there to support massive AI but no sign of it yet. The closest we have is ReadIris for scanning to digital or a Speech Recognition piece of crap software (from any company). I’ll have to do it myself, damn it!
By the way, talking about numbers…
Given:
25 teraflops = 25,000,000,000,000 instructions per second
Speed of light = 186,000 miles per second
1 mile = 5280 feet
A computer user sits 2 feet from the monitor
Then:
The Army’s Supercomputer will be able to perform 50,915 instructions in the time it takes the light to go from the monitor to the user’s eyes.
Or, to put it another way, the computer can average 1 instruction in the time it takes a photon to travel 1/2100th of an inch (.00047″, .01 mm).
Technically, it actually takes 200 times this long to do 1 instruction, but it can do over 200 instructions at once. Am I the only one amazed at this feat?
Waiting for the user to type a character must seem like an eternity to the computer.
…and the speed of light doesn’t seem so fast any more.
stop showing off lol
This cluster is optimized differently than the Va Tech cluster. The MACH 5 employs a larger number of nodes but with Gigabit ethernet interconnects . Clearly the focus is on batch processing CPU intensive tasks with less massive data transfer requirements. I liken this to the SETI@Home approach where is can take six hours to work through a 350KB data packet. The Big Mac has roughly 70% of the number of nodes as MACH 5, but uses infiniband interconnection.
I am not familiar with the testing methodology for the supercomputer cluster ranking, but Big Mac may be better in some of the tests and worse in others. It will be interesting to see the results. I will also be interested in how long it takes to set up MACH 5 and start performing useful work.
That information is freaking amazing Arugaetu!! thx for the info. well Apple is sure wowing the people who said they couldn’t compete with the wintel powers!
We just wanted to post our appreciation for the ironic humour of Sputnik’s many posts in various threads. And what’s even funnier is the response of the posters who totally don’t see it and think he’s trolling. Fantastic! Get a humour transplant guys!!
So when is somebody going to take the next step: spend $50 million, build 10 clusters, and cluster them together.
i do enjoy sput’s and one-nut’s commentary. but today is a lsow news day- thus the wit not top notch. maybe tomorrow!
See, I made a claim one time the Sput was a sarcastic smartass. And I got ‘corrected’.
Well I think he’s MDN’s ‘Don Rickles’. Either way, his comments are far to outrageous to be taken seriously.
Aryugaetu,
Just to be a stickler for accuracy
flop == FLoating-point OPeration (the original definition back in the 70s when a 12 MFLOP/s Cray-1 was considered extremely fast)
It has since become to be understood as
flop == FLoating-point OPeration per second
So really either is correct depending upon how far back you want to go.
Main point, however, is that it is specifically floating point operations, not instructions. The measure of instructions is MIPS (million instructions per second) or GIPS or TIPS. Of course there was the argument for years between IBM and DEC about how many instructions it took to perform a specific operation. The general consensus in the comunity back then was that it took two DEC VAX instructions (on average) to perform what one IBM 360/3090/etc. instruction did.
Also the PPC chip has a Multiply-Add-Fuse instruction which does several operations with one instruction.
This is why the supercomputing community tries to standardize on the floating point operations done. Who cares how many instructions or memory moves or NOPS (no-ops) are done in the process of getting the work done? What matters to them is how fast it can accurately calculate the final answer.
While I don’t believe the LINPAC benchmark is a great one (my personal favorite when I was actively in that field years ago was SLALOM), it is much better than counting the number of instructions performed.
— Just my 3 cents from an old hacker (not cracker).