Slashdot Mirror
Cray SX-6 Installed in Alaska
Dhrakar writes: "Now, I know that normally press releases are imediately round-filed, however, as this is the first NEC^H^H^HCray SX-6 to be installed in the U.S. it is newsworthy. The 8cpu, 64Gb system has been installed at the Arctic Region Supercomputing Center for benchmarking and other testing. See either ARSC or the NY Times (sub. required. Yada, yada) article."
Were they able to get a discount in not purchasing cooling equitment due to location? I suppose Alaska could be the paradise for heavy metal and overclocking...
Why bother.
Before anyone trolls about putting it in Alaska to save on air conditioning, Fairbanks gets into the 80F in the summer. Just thought I'd clear that up.
Maskirovka
Is a counter troll still a troll?
Yeah, it's supported by Veritas NetBackup DC already. That and my TI calculator and GBA.
I thought cray was dead, but it turns out, they were just using BSD.
Do you even lift?
These aren't the 'roids you're looking for.
what a waste! they should give it to me so i can play games on it! who cares about the weather anyway...
Just in case you want to play with toys like these, the ARSC is looking for an admin.
Learning HOW to think is more important than learning WHAT to think.
Cray SX-6 Installed at ARSC
Fairbanks, Alaska - The Arctic Region Supercomputing Center (ARSC) and Cray Inc. (Nasdaq NM: CRAY) announced today an agreement that places a Cray SX-6 at ARSC. ARSC is pleased to be able to offer this leading technology to the wi
Oh wait a minute, it's a f*cking supercomputer! Sorry about that.
What I am waiting for is the Cray SV2 which can have up to 1024 Cray vector processors. Who needs a beowulf cluster?
Computer Science is no more about computers than astronomy is about telescopes. --E. W. Dijkstra
Cray advertisement for the SX-6
Cray product sheet on the SX-6 (PDF).
A system that can pump out 64 gflops only running at a measly 500Mhz? Really shows how poorly mhz is as a measure of system performance.
It shall be used to create, download, store, and compile the WORLD'S MOST POWERFUL PORN.
hmm for all the people who wanna figure out what it would cost to run one of theese babies.
This link states in it that:
The "SX-6 Series" will be shipped from the end of December 2001 with the monthly rental price starting from 2,800,000 Yen.
By my calculations thats actually only about 22 thousand a month in dollars... not like im gonna be grabbin one, but frankly i would of thought they charge more
Tux would like it up there.
(sub. required. Yada, yada) - Well not quite THAT remote.. Personally I think alaska is TOO big of a cooling solution.
And my god these machines are beautiful and fast. You won't believe how much they can do. Of course, they're not as fast as the ones used for the nuclear simulations and stuff, but they make your AMDs and Intels look like horse and carriages compared to a Ferrari. I have the honour of building one of these machines. It sucks about 50kW of power. You can only dream of getting one of these machines.
They say 8 cpu's, 64 GB ram, 1 TB disk, 64 GFlops peak performance. That hardly sounds like a supercomputer by today's standards. A single processor AMD Athlon is capable of (I think) around 8 peak gigaflops (2 Ghz * 4 SIMD operations using SSE instructions). Similarly the 8 GB of RAM and 125 GB of disk per CPU is in midrange workstation territory. While there's probably a much higher bandwidth memory system than you could get out of an 8-16 node Athlon cluster, it's not clear what problems this Cray unit will really be used for that couldn't as easily be done with a rack full of PC's or workstations.
Check out the ARSC's website... they have some pretty snazzy hardware! SV1ex, few other Crays, several big SGIs...
.edu had that kind of money!
Wish my
The ARSC is well known for ordering their Crays in custom colors (usually white with black trim). They have some photos of their machine rooms on their website... the only white SV1 I've ever seen! Few other unnaturally white machines too!
In adolescence, where Farrah Fawcett should have graced my wall, there was a picture of a Cray SuperComputer in full splendor framing no mere mortal SysAd but a Dude who went by the name ArchAngel. Respledent all in White he and he alone touched the holy of holies. Now it's just dross for drunken /. trolls, oh my lost youth.
"Academicians are more likely to share each other's toothbrush than each other's nomenclature."
Cohen
- Find new prime numbers.
- Search for Intellegent life.
- Crack Crypto.
- Play Doom 3 on it.
Come on now, which one of these sounds the most entertaining?/^[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,4}$/i
Details are here
And yes, you get to play with the new Cray.
For more information, please contact:
Thanks! We're looking for someone with experience with supercomputers.
But the only thing you'll ever do is jack off to her picture. I would say that masturbation is pretty much the same thing as trolling.
I guess you've never masturbated to a picture of a Cray SX-6 before.
Aw, fuck it. Let's go bowling. - The Big Lebowski
What other industry can you get a job in Alaska or Hawaii doing the same thing? You might even end up inventing the next Mosaic out in the cornfields. Gotta love them pork-barrel politics!
When information is power, privacy is freedom.
Beside the fact there is no 2GHz Athlon, you forget one very important thing: memory bandwidth.
A usual Athlon has a theoretical memory performance of 2.1GB/s. Now do 8 gigaops on 32 bit float numbers. That would translate to 32GB/s. So 8 gigaops is not sustainable. Just a short burst.
And don't forget that that SX-6 has 2048 memory banks. Best Athlon chipsets I know have 1 (in words: one). Best Xeon chipsets have 2.
So while the raw power of supercomputers and PCs look similar on a sheet of paper (peak performance, AKA speed you can never exceed) supercomputers are built to get most of that performance not only for a short period of time.
Another topic is price/performance. Here a plain PC cluster might be better. But if you cannot parallelize a problem that much, one fast computer solves a problem faster.
agreed, I'm still laughing over that one.
odly enough
i believe the sega dreamcast pulls 1.4gflops to match the cray youd only neeed 46 dreamcasts which means for a measly 2300 you could pull 64.4 gigaflops much lower than 22k per month!
Where's the obligatory Beowolf Cluster comments?
how can it help you get a date???
Take the cheese to sickbay, the doctor should see it as soon as possible - B'Elanna Torres, "Learning Curve"
8 athlons would melt all the snow in alaska
(obpost)
Sounds like something apple would use in an advert
But if you cannot parallelize a problem that much, one fast computer solves a problem faster.
In bioinformatics, one of the more power-demanding applications of super computers, there are many problems that can not easily be split up in smaller independent pieces. 32-bit memory addressing is often a problem as well. Of course these problems can be circumvented, but in the end it all comes down to speed and not having to re-engineer complicated scientific code.
IANAL, but imagine a beowulf cluster of in Soviet Russia all your belong are base to us welcoming the new SCO overlords.
The Supercomputing Center for benchmarking and other testing? "Other" huh?
Can this be some new hardware for the National Missile Defense that Bush is building over in Alaska?
If you have to run applications, where you can not make much use of vectorized instructions, then these systems are not faster than any other computer is, too.
There are two american companies which are developing very impressive technologies:
- IBM tries to build hypercomputers (quantum computers), and research results look pretty promising. - a few images
- They are going to build a One-Petaflop Supercomputer until 2005 utilizing 1,048,576 Processors (32 Cores per Chip, 64 CPUs per Board, 8 Boards per Frame, in 64 Frames) - Blue Gene Project
- They are developing CPU Cores, where all execution units are connected asynchronously - that makes it easy to reach an extremly high clock frequency.
Speed per processor doesn't matter - just think about Intel SMP systems compared to RISC SMPs. Scalability is the one thing that matters in supercomputer technology.
The 8cpu, 64Gb system has been installed at the Arctic Region Supercomputing Center for benchmarking and other testing
2847 in Content Creation Winstone.
3000 in Business Winstone.
Ok, pack it up. Next!
I would say that masturbation is pretty much the same thing as trolling.
Another step towards the Grand Unified Theory of Slashdot! An AC has managed to link the forces of self-abuse and the abuse of others into one, simplifying the universe.
I agree that Americans need a more heterogeneous set of supercomputers these days. Vector computing has "gone out of style", but it's still very very useful for a lot of applications. We may see vectors return somewhat with this reselling plan, and with the soon-to-be-released Cray SV2.
It can, but the matches would be SO precice as to eliminate all feeling of romantic conquest.
-Prof John Frink
How about a rack full of dual processor anthlons? Oh - that is not one computer? Oh - sorry - you draw the boundries where you want but when all machines are running the same 3-D geophysical migration it seems to me that they are one machine.
I'm not impressed. I'll bet that the anthlon rack will compute circles around that cray and cost far less. Not only this, individual units can be pulled and fixed or replaced rather easily.
I'm reading down further at the comments about comparing the stinky desktop PC to a "super computer" and I have to chuckle at the ignorance. The company I'm thinking of that put the anthlon rack in place for the 3-D migrations had an Alexis (sp) then about 100 sparc's networked. As one of the bigger geophysical processing shops in Calgary and Houston I rather think that they know what they are doing.
Er, don't you think that they would take the time and money to rewrite the applications to use the vector processing thingies? I mean, they're paying a lot of money to use these machines, so it would make sense that they'd shell out a few thousand dollars on the side to make their programs work best.
Take Motorola's Altivec, for example. Apple wrote a bazillion lines of new code to take advantage of Altivec. Yes, Apple could have just stuck with the G3, yes, Apple could have saved a lot in terms of paying programmers, but scrimping on coders means that you don't have the acceleration code necessary to use the CPU to its full potential.
Gah! What I'm trying to say is that when you shell out a lot of $$$ for a computer, you usually try and make the most of your money.
I will now redundantly add my name to the end of my post. You know, in case you forgot me or something.
So >that's what's melting the glaciers.
it gets about 923749083274fps in quake III
i'm amazed that i survived - an airbag saved my life.
It's NOT a measure of system performance.
It's a measure of clock speed.
It's like saying "This engine tops out at 2000rpm, but this engine here can do 4000rpm"
Is the second a more powerful engine? Hardly.. the first is out of a huge diesel caterpillar; the second is out of 20 year old Honda Civic.
Yeah. But can they sustain the same performance?
What about price/performance?
Alaska's temperature has risen 9 degrees in the past century. Why the hell are they installing super computers there? Maybe they should put their heat transfer unit inside a glacier.
(I know it's inisignificant amount of heat increase, but still... May be a start of a trend?)
(-84 farenheit to +102 farenheit according to another account)
They were on the news a while ago with kids not having shoes (in winter!) because of the financial situation. When it gets below -60 degrees C, some of the equipment stops working (similar to the situation in the UK when it gets below -1 degrees C).
And they have really nasty floods there too.
Is this show going to go Fairbanks and help the super computing geeks find hot dates with hot ski bunny types?!
--
Billy Corgan: Billy Corgan, Smashing Pumpkins.
Homer Simpson: Homer Simpson, smiling politely.
"Unlike most of you, I am not a nut." - Homer J. Simpson
People *seriously* underestimate just how pathetic the memory bandwidth is on your standard desktop PC.
For the coders among you: Suppose you had an algebraic structure datatype that you had test against a set of n! permutations. Standard programming dogma says: Generate the permutations once, store them in memory, and then grab them as needed... right?
At least on my Athlon XP (and, I suspect, any modern processor with a piece of crap bus)... WRONG. It ends up being MUCH faster to regenerate the permutations from scratch every freaking time you need them, rather than risk having a cache miss and grabbing them from RAM.
I know you won't believe me, because I didn't believe me at first either. I couldn't imagine that the memory bandwidth was THAT BAD. I coded it up this way to see how much WORSE it performed... and it ended up performing better. An important lesson about optimizing programs for modern Intel/AMD architectures was learned: often times is faster to recompute on the 2GHz processor, rather than wait for the not_2GHz_bus to fetch information from RAM.
But please, don't take my word for it, go try it for yourself.
Because....
The memory bandwidth of E10k's is a rounding error compared to an SX-6...
and... SPARCS aren't vector processors.
But since you think CPUs + RAM == net performance of a computer, I can safely assume you probably haven't the foggiest idea what a vector processor is, or how one could take advantage of it.
And its not a Cray anymore than the Dodge Stealth was truely a Dodge... the SX-6 is made by NEC and re-badged as a Cray for sale in NA.
well, either that, or with so much processing power, they are up to a plan to summon the devil himself and rule us all ! :D
The politics that follow this 'sale' ought to be rather interesting. NCAR bought a Japanese supercomputer some time back and nearly got wiped out by funding deletion by the US Congress.
What happens next ought to be VERY interesting.
On the other hand, the Cray employees I've talked to - needling them for giving into the dark side and selling a SX-6 - have said that anything that is good for vector computing is good for Cray: they can always sell a follow-ob with their SV-2 and SV-2e.
I saw a post that I skimmed above that stated something to the effect that "you'll never touch [a supercomputer]. We, at NERSC, are still looking for a few good sysadmins. Keep in mind we're pretty brutal about who we let in, but if you think you have the right stuff to be a sysadmin on some of the world's most powerful machines...;)
Do you know why the road less traveled by is littered with the bones of the unwary?
Mainframes often have several hundred MB (or maybe several GB by now) of SRAM (20 ns latency or so) along with many GB of DRAM. If this 64 GB on the Cray is SRAM that's more impressive. But even SRAM (20 ns is 40 cycles access time) is orders of magnitude slower than on-chip cache memory (1-2 cycles). So the Cray has the same locality issues as a PC.
One VR helmet: $5000
One VR mouse: $1000
Integration cost of the VR equipment to the CRAY: $25,000 (roughly)
Spending the rest the next 12 months with your VR girlfriend in true cyberspace: Priceless.
There are things you can buy, and there are things you can build, and then there are things you can get buy building and buying and having a perverted imagination.
I am not really sure what the spec is on the cray, but just trying to imagine a cluster of atholons trying to access 64GB of non-uniform memory, across network latency, bus, and then through the memory sub-system, and finally to the main cpu that is trying to read and write to the same memory location as that of the other god knows how many other cpus (assuming a cluster of 240 nodes), and the software complexity to manage all that memory, not to mention to manage the cache coherency, making it like a CC-NUMA system, which using standard pc components rely on very complex software to provide an interface just to provide memory management, it is already quite complex.
:-)
With the cray, with less cpus to deal with and bigger foot print of main memory, each cpu would have more to work with where as the cluster would have less to work with per node.
Another thing to consider is that these are vector processors which already have a solid base of development for weather simulations, nuclear bomb testings, and such ungodly application usage. (Which is why the PS2 is treated as a munitons because it too is also a vector processor.)
I am not writing this to put down the work done for in the area of beowolf clusters and the like. But you have to look at the application that is being used for and what they intend to do with it.
Another thing to consider is why should you try to get a x86 processor to do vector processing? It is like when cyrix tried to do floating point instructions a few years ago in software because it didn't want to put one in. It could never outperform a FPU that could do floating point calculations on hardware. In order for you to do vector processing, you would need to do what the cray does on hardware emulated in software. Just might not work... maybe in transmeta though.
Anyone have thoughts on this?
Too bad somewhere along the way we lost the American Way.
Seastead this.
We have several systems at work with up to 24 gigs of RAM and 18 CPUs. Why is the installation of this thing that important?
If it had 64 terabytes of RAM, it'd be interesting.
- A.P.
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
(up to) 1,024 Processors....@ 500Mhz *EACH*
...so thats the equivelent of 0.5Thz =D
or
500-fucking-GIGAHERTZ!!!
They are developing CPU Cores, where all execution units are connected asynchronously - that makes it easy to reach an extremly high clock frequency.
Where I come from asynchronous means without clock. So as much as you would like to believe, these processors will have no clocking frequency, but you may say that they will run much faster (assuming good design) compared to clocked processors.
They were supposed to sell a number of SX machines since they struck the deal with NEC. This is the *only* one they managed to sell, and since they didn't have much money left to sustain their own SX-6 installation in Chippewa Falls (which was the actual first US SX-6 installation), they sold their machine to the ARSC, with a deal that lets them use it sometimes for testing and training.
My feeling is that they are utterly uninterested in selling SX systems, they'd rather sell their more profitable SV systems or their crazy MTA systems (woohoo, they managed to build ONE).
Disclaimer: I used to work for Cray in their SX support team and with HNSX Supercomputers before that, the North American subsidiary of NEC for supercomputers. I left of my own will, I wasn't part of the friday-right-before-Christmas round of layoff Cray did.