Cray supercomputer is world's fastest (that we're allowed to know about)
By Tim Stevens 
posted Nov 12th 2008 8:50AM
posted Nov 12th 2008 8:50AM
Seattle-based Cray has been wowing us with massive, ridiculously fast supercomputers since the '70s, establishing a position for its machines high on every geek's most wanted list -- despite never publishing a Doom benchmark. In recent years the title of "world's fastest" supercomputer had been lost by the company, ping-ponging from Wako, Japan to Armonk, NY, but is now back in Cray's hands with the implementation of the XT Jaguar. It's comprised of over 45,000 quad-core Opteron processors, 362GBTB of memory, and has a 10PB (petabyte) storage array, able to perform calculations at a massive 1.64 petaflops. That's over one and a half quadrillion operations per second and more than 50 percent faster than IBM's previous heavyweight. Mind you, Cray is quick to point out that this is the fastest machine being used for non-classified research, a caveat that just makes us even more curious about what's keeping the Pentagon's server rooms warm and loud... and apparently orange.
Update: That should be 362 terabytes of memory, not mere gigabytes.
[Via UPI.com]
Update: That should be 362 terabytes of memory, not mere gigabytes.
[Via UPI.com]
I need one ;)
I need two.
With the rise of GPU over CPU superiority for specific task, it will be interesting how the super computer will take shape in the future.
Read somewhere that transcoding using nVidia 260 GPU (reference: Badaboom and Protein Folding) is about 20 times faster than Intel Quad. So does this mean 4500 opteron Quad is about as powerful as 225 nVidia 260 GPU ? (busy finding SLI interlink connector..hehehe).
The party will get merrier when intel unveil Larrabee GPU (cough), easy programmable in C+ and C++ (correct me if i am wrong) .Hear that ATI, nVidia? So when Direct X 12 released we just need to replace the driver with Larrabee (cough) instead of replacing the old video card - provided its still fast enough).
Only time will tell (yeah.. real soon)
@vampritt
Faster, yes. But also less precise. Many other network/cluster computing projects haven't adopted GPGPU exactly because of this reason.
"Specific task" is indeed the keyword here. There will be no universal take-over.
I need to destroy this. Its starting to sound a like like skynet......
I don't even want to know what the classified ones are capable of....
sorry folding is Stanford Univ not MIT (blushing)..
check this out guys... it already happed!!!!
http://www.gpgpu.org/
GPU is taking over specific and general purpose computing.
But can it play Doom??
45,000 quad cores wtf? best hope theres never some kind of shortage, or we will have to stop this nonsense.
*grumbles* 45,000
? You're acting like this is someone's PC... It's a bloody super computer that will be crunching numbers for various projects 24/7.
will they be releasing a netbook sized version of this anytime soon?
*thinks* no:P
Yes.
In about 50 years or so.
Not even 50 years, judging by Moore's Law I'd say you could expect these specs in a laptop some time around 2032 (if laptops still exist then.)
yeah, but MS won't let them sell it with XP.
So is the name of the company Crazy or Cray? I know its early and all...
362 GByte? I guess TByte.
It better be TByte. 362GByte would mean a mere 2MByte of RAM per core. Hardly believable, given that each core probably has more cache than that to work with.
quite possible that it's only 2mb/core, it's meant for highly parallel computing anyway, so no core needs more than a small portion of the overall picture. In fact, chances are that most cores have far less memory to work with and a few hubs take up the majority of the memory.
someones probably already mentioned this, but when i saw the image, i thought 'what does Cray have to do with the car compnay Jaguar?'
Can i get a discount for bulk orders?
.. building a beowulf cluster?
Freakin' Sweet
Hate me if you must....but, Will it play Crysis?
It played it, and beat it 34,000,000,000 times in the past 55 seconds.
Sod Crysis, will it run Vista? ;)
@Steve
It RAN Vista RAN Windows 7 chewed up and spit out Leopard owned Linux then realized they were all not robust enough contemplated and created Windows 2015. All before calculating 42.
Well... in all likelihood, we will never know. I doubt that anyone ever really tried running anything else than serious work on a large-scale cluster computer. For one very particular reason.
screw Crysis... this monster can divide by zero and... dare I say it... succeed! *gasp*
I heard the classified ones are fast enough to finish an infinite loop in 8 minutes.
Pssh, my mind can run circles around that thing.
This thing can run SKYNET for fucks sakes...
I personally dub it the
Cray Engine.
Typo: Aramonk should be Armonk
Or how the troll should be under his bridge.
And WHY hasn't iEye been hit with the ban hammer yet?
Jesus, these people (iEye and the like) only post to get people's reactions and attention. Don't even read the comment, just low rank his ass and then maybe he'll get bored and fuck off. 12-year olds generally get bored pretty quick I thought lol.
Anyway, about the computer. Simply amazing, I wonder how much one of these bad boys costs? Anyone know any one who has one for pre-order, we could all go around and gawp at it haha.
In some years we'll laugh about how our netbook once was huge.
Devices can only get so much smaller and still be usable...we've already reached that point with cell phones and we're knocking on the door with the latest netbooks.
The screen size is a big limiting factor - they're not going to get any smaller because there's a useful limit - until *cheap* flexible OLED screens that can be rolled up come into play (i assume). That will dictate the area mostly, the depth will mostly be dictated by having a tactile keyboard - and of course the power of the book. The limiting factor of nm then comes into play with chips, we can only go so small before we have to start messing with things like taking quantum effects into account - and if Intel keep at it then we're going to hit that wall in the near future. Of course with things like 3D transistors and quantum computing on the horizon, we won't have to worry about that ;) And of course nanotubes will be in everything..
The power, rather than size is what we'll be commenting about in the future. The average netbook knocks the socks off most computers built over 5-6 years ago.
Can it run C....
oh well ....
Can I be highest rank too?...please
lol...I was supposed to reply to the comment below.
I look like an idiot don't I
To put a weird angle on this: it takes about 100 teraflops to simulate a human brain down to the neuron level, so you could say this thing has a brain 16 times the size of a human's. Eat your heart out, Einstein!
Thats fine, but try asking it if it likes toast or bagels. Instant BSOD :) .
Betcha you can shut it down with some old-skool tic-tac-toe.
It is the software, not the hardware holding AI back. And it has been for a long time. Software is not following Moore's law.
Greetings Professor Falken, shall we play a game?
How about a game of Global Thermonuclear War?
That is an incredible simplification... You can't directly compare large arrays of conventional processors with the potential information processing capacity of the human brain by simply comparing theoretical peak FLOPs. In fact, I don't believe we are even close to understanding the intricacies of neural processing. I can't remember the details, but recently there was an article about a research project where a modern supercomputer was able to simulate a tiny slice of rat brain at a fundamental physiological level, and it took like weeks of processing to simulate something like 100 nanoseconds of time...
Like the teasers for Red cameras, I'm getting increasingly bored of these announcements. Forgive me if you disagree, but it's hardly amazing anymore, because all you do to claim the crown is throw more and more processors at it. The only amazing thing is that someone has stumped up enough cash to get that many chips hooked up together.
If I had a million-billion-trillion pounds, I could just link up several XT Jaguars, and declare myself as owning the fastest supercomputer. Hardly amazing, is it.
I used to be wowed by supercomputer stats, when I was 15, but now I guess I'm old and cynical and can see that all it amounts to is large sums money rather than pure technology.
Like Red cameras, I still want one though.
"If I had a million-billion-trillion pounds, I could just link up several XT Jaguars, and declare myself as owning the fastest supercomputer."
So, the answer to the question at the bottom of the article...
The "million-billion-trillion pounds" might not be 'accurate', and it would also depend on your personal definition of "several"
Good point. It used to be that supercomputers were at the leading edge of technology, but now they are basically large numbers of PC chips soldered together.
@ Labrador
That's exactly my point - in their heyday, supercomputers consisted of fairly custom processors, and probably only a handful, each of which were far more powerful than anything else in the world. Now, as you rightly mention, it's just a load of cheap chips, but masses of them.
Yes, you might get better performance per dollar that way, but the romance and intrigue has been lost because of it.
It's not *just* the number of processors, though. Isn't there some pretty clever architecture around connecting them all, getting them all to run the right programs at the right time, etc?
That said, ultimately you need the clever architecture whether you have a top500-leading thing or something 'slow' with half the processors, so yes it's all about money as well.
Those computers crunching away on Classified programs are known as Blue Gene and ASCI.
The Pentagon has 5 of these, 1 lining each wall.
I'll betcha Batman has two of these.
Like Red cameras... the question is, will it - blend?
(scoff) Cores? Petaflops?
I won't even get out of bed for anything less than Spacely Sprocket-based.
Cray should join Engadget's folding@home team. Stick it to PS3Fanboy's f@h team.
I'm waiting for the flexi-screen XPS version.
You know that they cannot store files on this petabyte computer. It's against the law to harbor petafiles.
@Larry
Hilarious, I know you had to be waiting to use that one, but very funny none the less.
Lawlzz
Can I haz a supercomputer :P
If your going to make one of those "I can haz cheezburger" posts at least do it right.
you gotta wonder what kind of an operating system this thing runs, damn sure its not vista, or os x
then is must be Ubuntu.
It runs it own made OS. Just the necessary to be able to execute its calculation programs...
that and Firefox with a Add-on to download all the internetz
It runs a version of Scientific Linux. I can't remember the version number.
IIRC it's a stripped down SuSE Linux.
will this play crysis?
Why you guys ask this question a lot I can run Crysis on my core 2 duo PC. Why this wont run CRYSIS! it can run 45000 Crysis the same time !!!!
3Dman just got trolled
@3dman:
sar⋅casm
/ˈsɑrkæzəm/ Show Spelled Pronunciation [sahr-kaz-uhm] Show IPA Pronunciation
–noun
1. harsh or bitter derision or irony.
2. a sharply ironical taunt; sneering or cutting remark: a review full of sarcasms.
Thanks!
I have my own dictionary.
@ Larry
Hahahahahahahahaha!!!!!!
That's what they said 35 years ago about quadrophonic.
So how many FLOPS does a regular 8800GTS (or GTX or 280GTX what ever) calculate?
Also... What kind of stuff does this thing calculate?
Any reason the IRS doesn't lease this bad-boy at tax-time?
They would crank out the returns too fast and lose money or something...
"It's comprised of over 45,000 quad-core Opteron processors..."
Should be "composed of."
The thing I hate most about all this IT stuff is that in a few years time we will all be joking about how our Palmtops have more compute power than Crays super computer had back in 2008.
My 100 watt stereo amp, on the other hand, will still be kicking in 20-30 years time...
In 20-30 years time, 'stereo' will be so old-school!
If you want to see what this system actually looks like check out http://www.nccs.gov/jaguar/ Engadget shows the Cray XT4 which while quick isn't as quick as the XT5 which makes up the majority of this computer. Also check out the time lapse video to see them installing all 200+ cabinets.
180,000 CPUs .... wow
I use to work as Net. Engineer for ISP and we had something similar yet not that powerful. More memory for sure! but less storage. We worked with Broadbus.One box nix* enterprise edition replaced 78 servers! Uptime when I left 491 days :) no problem no glitches.
Cost 1.5 mil :)
Now owned by motorola this is what I use to work with:
http://www.motorola.com/business/US-EN/B-1_US-EN.do?vgnextoid=1abf9ffbede46110VgnVCM1000008406b00aRCRD
LC~
http://www.livecrunch.com
too bad Chuck Norris is still faster
IBM's supercomputer dubbed Blue Gene/P is designed to eventually operate at THREE petaFLOPS.
nVidia GeForce 8800 GTS 512MB gets 416 GFLOPS. I believe the 280 along with ATi 4870 have broke 1 Tflop, but just barely. like 1,200 Gflops.
but they are talking about the CPU calculations, not GPU. they do different calculations, and GPUs will get a higher number of FLOPS on a rating, but they can't perform the same operations a CPU can, so they are not really comparable.
as for CPUs, the new Intel Core i7-965 Extreme Edition @ 3.73 GHZ gets 69GFLOPS, the AMD Phenom Black 9550 gets 29GFLOPs - compared to 1,640 GFlops (1.64PFlops) - so the cray is roughly 24x faster than the fastest processor out.
http://mos.techradar.com/techradar-corei7-benchmarks.pdf
is that all right?
I b el
@kyle (are u the famous kyle_brenet from HardOcp?)
Yep. Specific versus general purpose. The current reprograming barrier will hinder adoption to CUDA and other language utilising GPU processing power. However once the foundation is laid out, there is no turning back. Something like converting video format to another (transcoding). Will you use your latest shiny Quad Core to transcode H.264 to MKV while your videocard can do it 20 times faster? Or use your processor calculate something complex while your videocard can do it 20x faster? http://www.hardfolding.com/
Maybe media convertion and GPU power to find the cure for cancer is all they can think right now, but the list is growing soon. Why not?
@MarbleMind
Less accurate for GPGPU calculation? hm.. then why the hell the MIT used ATI and nVidia videocard to calculate the Protein folding along with PS3 cell processor? Its better to ask them why. But in my humble opinion, its not about accuracy. - sorry dont get offended...i might be wrong..
Remember USB legacy? Intel is placing all the burden to CPU because its their bread and butter. They need everything to be depended on CPU and are very uncomfortable that GPU will take over specific function of the microprocessor that buying latest fastest new processor will be second priority. (No more buying latest $1000 bucks processor to play the latest game at hyper speed, instead, spend only $299 bucks on latest videocard.) If this thing goes on, CPU will reduced in function just as its fast enough to handle general purpose and fast enough to feed data to the GPU to process. math co-processor anyone?
For the super computer, since most of the time, it will sit there and crunch number until its done and will wait for the next variables to calculate. (that is why they have list of schedule on what the supercomputer need to calculate after they are done with the tasks given). Dont you think GPU is much suited for this job? ---> Supercomputer = specific task. Heck maybe I foresee the future of supercomputer will use 45,000 nVidia and ATI processing chip. (er.. if each card get their own 1GB memory, it will be hell of a memory = 45TB) heheheh.
Maybe with that kind of processing power, we could resurrect Nikola Tesla's free energy, cold fusion, Stanley Meyer's Water power or maybe just sit there and play Crysis 3 on the notebook version of the system in 3 years time. Take a Vote Guys....heheheh
Hear that?......... The wind of change is blowing at the CPU GPU battleground.......
Sorry if i gargle too much.. :P
@kyle (are u the famous kyle_brenet from HardOcp?)
Yep. Specific versus general purpose. The current reprograming barrier will hinder adoption to CUDA and other language utilising GPU processing power. However once the foundation is laid out, there is no turning back. Something like converting video format to another (transcoding). Will you use your latest shiny Quad Core to transcode H.264 to MKV while your videocard can do it 20 times faster? Or use your processor calculate something complex while your videocard can do it 20x faster? http://www.hardfolding.com/
Maybe media convertion and GPU power to find the cure for cancer is all they can think right now, but the list is growing soon. Why not?
@MarbleMind
Less accurate for GPGPU calculation? hm.. then why the hell the MIT used ATI and nVidia videocard to calculate the Protein folding along with PS3 cell processor? Its better to ask them why. But in my humble opinion, its not about accuracy. - sorry dont get offended...i might be wrong..
Remember USB legacy? Intel is placing all the burden to CPU because its their bread and butter. They need everything to be depended on CPU and are very uncomfortable that GPU will take over specific function of the microprocessor that buying latest fastest new processor will be second priority. (No more buying latest $1000 bucks processor to play the latest game at hyper speed, instead, spend only $299 bucks on latest videocard.) If this thing goes on, CPU will reduced in function just as its fast enough to handle general purpose and fast enough to feed data to the GPU to process. math co-processor anyone?
For the super computer, since most of the time, it will sit there and crunch number until its done and will wait for the next variables to calculate. (that is why they have list of schedule on what the supercomputer need to calculate after they are done with the tasks given). Dont you think GPU is much suited for this job? ---> Supercomputer = specific task. Heck maybe I foresee the future of supercomputer will use 45,000 nVidia and ATI processing chip. (er.. if each card get their own 1GB memory, it will be hell of a memory = 45TB) heheheh.
Maybe with that kind of processing power, we could resurrect Nikola Tesla's free energy, cold fusion, Stanley Meyer's Water power or maybe just sit there and play Crysis 3 on the notebook version of the system in 3 years time. Take a Vote Guys....heheheh
Hear that?......... The wind of change is blowing at the CPU GPU battleground.......
Sorry if i gargle too much.. :P
Way to say where it is installed and what it is doing. (which by the way is Oak Ridge National Lab and research into superconductors.)
The real question is how much money are we spending in taxes to keep this thing running
We all pay taxes. Get over it.
That's all well and good, But the important question is ... Can it play Lock-On Smoothly? lol
"you gotta wonder what kind of an operating system this thing runs, damn sure its not vista, or os x"
It runs compute node linux, a stripped down variant of linux (SuSE linux I believe).
Geeks are not funny, so stop making jokes.
...45,000 quad-core Opteron processors, 362 TB of memory, and a 10PB (petabyte) storage array... and an 8" screen.
Coming soon to a netbook near you!
yeah but can It predict the next engadget meme?
OOOOOOH. Let's use this to crunch numbers, create credit derivatives and make mortgage-backed securities! hahahaha