Beta
×

Welcome to the Slashdot Beta site -- learn more here. Use the link in the footer or click here to return to the Classic version of Slashdot.

Thank you!

Before you choose to head back to the Classic look of the site, we'd appreciate it if you share your thoughts on the Beta; your feedback is what drives our ongoing development.

Beta is different and we value you taking the time to try it out. Please take a look at the changes we've made in Beta and  learn more about it. Thanks for reading, and for making the site better!

Titan Tops Top500 Supercomputing List

samzenpus posted about 2 years ago | from the top-of-the-heap dept.

Supercomputing 52

miller60 writes "The new Top500 list of the world's most powerful supercomputers is out, and the new champion is Titan, the new and improved system that previously ruled the Top500 as Jaguar. Oak Ridge Labs' Titan knocked Livermore Labs' Sequoia system out of the top spot, with a Linpack benchmark of more than 17 petaflops. Check out the full list, or an illustrated guide to the top 10."

cancel ×

52 comments

Sorry! There are no comments related to the filter you selected.

I'm looking forward to the.. (4, Funny)

mrbluze (1034940) | about 2 years ago | (#41959345)

2013 Calendar full of naked supercomputers displaying their petaflops!

Obligatory (1, Funny)

Andrio (2580551) | about 2 years ago | (#41959353)

Obligatory "Crysis Max Settings" joke.

Re:Obligatory (0)

Anonymous Coward | about 2 years ago | (#41959421)

Obligatory "Still getting shitty FPS" joke

Re:Obligatory (1)

Anonymous Coward | about 2 years ago | (#41959643)

Missed. Titan has NVIDIA GPU's.

Re:Obligatory (4, Informative)

cashman73 (855518) | about 2 years ago | (#41960775)

This is significant, because if you look at the list, Titan only has about 500,000 cores, versus 1.5 million cores on Sequoia. The overall increase in peta-FLOPS isn't that much, but the fact that Titan did this with only about 1/3 the cores as Sequoia IS significant.

Counting artifacts... (1)

Anubis350 (772791) | about 2 years ago | (#41963943)

Bear in mind, part of it has to do with *how* cores are counted. I wasn't at the top500 announcement this year (didnt have the time to head to SLC for a week, wish I did), so I can't speak to if they discussed it, but at previous SC top500 announcements in the past few years they've stressed the difficulty in reconciling core counts between GPUs and CPUs, including (last I heard) settling on a 4:1 GPU vector procs:cpu equivalent ratio for NVIDIA cards.

Re:Obligatory (1)

fatphil (181876) | about 2 years ago | (#41965123)

I'm not sure the number of cores is reliable. One figure said a quarter of a million nvidea cores. That works out at 100 boards (2.5kcores/board). Another figure from an other article
http://www.anandtech.com/show/6421/inside-the-titan-supercomputer-299k-amd-x86-cores-and-186k-nvidia-gpu-cores
claimed Titan was 18k opteron with 18k nvideas, one each. The latter implies 50 million of nvidea cores.

As the latter has a photo of a rack, with 4 opterons (16 cores each) plus 4 nvideas (2688 cores each), I'm much more likely to believe Anand than top500.org.

Re:Obligatory (0)

Anonymous Coward | about 2 years ago | (#41965409)

Indeed, but on performance per watt, they are almost equal (the first needs a bit more power than the second). The CPU cores of the BGQ are probably more flexible than the GPU cores when the problem at hand is harder to vectorize. None of these is running at stellar clock rates.
However there are other interesting parameters which are not shown: I believe that the IBM CPUs are on an older process
than the nVidia GPUs, so it would be easier to improve performace/watt or increase clock rate with a process shrink.

Re:Obligatory (0)

Anonymous Coward | about 2 years ago | (#41959429)

I'll save everyone the trouble:

Imagine blending a Beowulf cluster of these while playing Crysis and Doom. That Cray.

Re:Obligatory (1)

sortius_nod (1080919) | about 2 years ago | (#41959887)

You forgot the hot grits & Natalie Portman!

Re:Obligatory (1)

Anonymous Coward | about 2 years ago | (#41959431)

Obligatory "beowulf" joke.

Re:Obligatory (0)

Anonymous Coward | about 2 years ago | (#41962895)

And the very best thing is it runs Windows 8 RT on an ARM emulator.

My previous place of employment - #71 (3, Informative)

Cutting_Crew (708624) | about 2 years ago | (#41959427)

NASA/Stennis Space Center, Mississippi. Used to be in the top 10. they have fallen way behind.

Re:My previous place of employment - #71 (0)

Anonymous Coward | about 2 years ago | (#41959469)

i love how they have a super computer, and are still behind. what an awesome world we live in

Re:My previous place of employment - #71 (1)

Anonymous Coward | about 2 years ago | (#41959675)

look at it this way: if they hadn't split their acquisition into 2 separate 391 (Rpeak) TFLOPS systems, they'd probably be good enough for #36.

Redudundant expression (4, Informative)

Anonymous Coward | about 2 years ago | (#41959461)

Sorry to be pedantic, but "petaflops a second" is redundant -- FLOPS means "floating-point operations per second."

Re:Redudundant expression (0)

Anonymous Coward | about 2 years ago | (#41959641)

Like sending a SMS. It's quite hard sending the whole "Short Message Service".

Re:Redudundant expression (1)

mozumder (178398) | about 2 years ago | (#41959849)

Maybe they mean the number of anti-fur ad campaigns that somehow backfired per second?

Re:Redudundant expression (1)

rubycodez (864176) | about 2 years ago | (#41959933)

thank goodness for that, I enjoy the 70s porn

Re:Redudundant expression (1)

psyhofreak (795109) | about 2 years ago | (#41960009)

I suspect that miller60 enjoys driving his car at 60 MPH per hour.

Re:Redudundant expression (1)

Anonymous Coward | about 2 years ago | (#41960043)

Sorry to be pedantic, but "petaflops a second" is redundant -- FLOPS means "floating-point operations per second."

Thank you so much. It grated on me so much, I came here to whinge about it. Now instead of being a complainant, I can merely congratulate you and look good. :)

Re:Redudundant expression (0)

Anonymous Coward | about 2 years ago | (#41963979)

Sorry to ruin your day, but the old FLOPS from the 70's is now written flop/s where "flop" means FLoating-point OPeration. Writing "17 petaflops a second" is correct.

Re:Redudundant expression (1)

froggymana (1896008) | about 2 years ago | (#41960525)

Maybe they're measuring the acceleration of it's floating point operations!

Re:Redudundant expression (1)

Latentius (2557506) | about 2 years ago | (#41962047)

Man, if it's operating at 17 petaflops per second, just think how fast it'll be by this time next year!

Re:Redudundant expression (0)

tbird81 (946205) | about 2 years ago | (#41961673)

Go easy on the author. This is politics discussion site, so not all of us understand basic computing terms.

Re:Redudundant expression (0)

Anonymous Coward | about 2 years ago | (#41973975)

Is it redundant? I thought a FLOP was a floating point operation. For example, an algorithm may need 100x10^15 Floating point operations, or 100 petaflops, to complete, but could take different lengths of time to compute depending on the number of floating point operations per second.

That's some accelleration... (4, Funny)

eth1 (94901) | about 2 years ago | (#41959465)

more than 17 petaflops a second.

Wait... 17 petaflop per second per second?! How long can it keep that up?

Re:That's some accelleration... (2)

stackOVFL (1791898) | about 2 years ago | (#41959687)

For a fortnight per furlong

Re:That's some accelleration... (1)

Alwin Henseler (640539) | about 2 years ago | (#41959733)

About 1.59 seconds, until it reaches 27.113 petaflops? Then it should decelerate about 0.56 seconds, stabilizing at 17.59 petaflops.

0 to max in 1.6 seconds, sounds like an awfully fast car^H^H^H machine.

Re:That's some accelleration... (0)

Anonymous Coward | about 2 years ago | (#41959745)

Exactly one second.

Petaflops a second? (1)

afidel (530433) | about 2 years ago | (#41959477)

Quadrillions of operations per second per second? So these machines only do scalar operations? I thought Linpack was a matrix test =)

But... (0)

Anonymous Coward | about 2 years ago | (#41959481)

does it run Linux?

Re:But... (0)

Anonymous Coward | about 2 years ago | (#41959553)

Yes, and the warmed coolant makes excellent hot grits.

Re:But... (1)

dzelenka (630044) | about 2 years ago | (#41961975)

does it run Linux?

If you follow the links in the article you will see that they all run Linux, or a Linux variant.

But how useful is it for REAL science? (0)

Anonymous Coward | about 2 years ago | (#41959607)

Petaflops mean nothing! They're just gaming Linpack! This new supercomputer has no capacity for doing REAL science! And oh... errr it's a US machine? Never mind then. USA! USA!

Shoehorn some actual work (0)

Anonymous Coward | about 2 years ago | (#41959777)

Between all the time it takes to upgrade these beasts, and the time spent running proper benchmarks, how much time is available to run actual jobs?

Re:Shoehorn some actual work (1)

drkim (1559875) | about 2 years ago | (#41959963)

Between all the time it takes to upgrade these beasts, and the time spent running proper benchmarks, how much time is available to run actual jobs?

Wadda ya mean, "actual jobs?"

They just built this to run Crysis 3.

Re:Shoehorn some actual work (0)

Anonymous Coward | about 2 years ago | (#41960179)

Running linpack is fast. There are machines on this list, that, besides being announced before, are running on the current capability for less than two weeks.

Petaflops a second? (1)

JavaBear (9872) | about 2 years ago | (#41960199)

"with a Linpack benchmark of more than 17 petaflops a second"

Supercomputers calculate on an accelerating performance curve now?

Re:Petaflops a second? (0)

Anonymous Coward | about 2 years ago | (#41960537)

They figured the only way to simulate a big bang, was to become one.

Re:Petaflops a second? (1)

gnasher719 (869701) | about 2 years ago | (#41961491)

"with a Linpack benchmark of more than 17 petaflops a second"

Supercomputers calculate on an accelerating performance curve now?

Sure. Just the number is wrong. If the top computer does 17 petaflops, and the list is run for about 30 years = about 1 gigasecond, then the average growth is 17 megaflops per second. Actual growth was a lot more in the last year, and less before.

And The software is??? (0)

Anonymous Coward | about 2 years ago | (#41960321)

When will these press releases start giving some information about the OS and the way of scheduling and patitioning the tasks?
pgmer6809

Re:And The software is??? (3, Informative)

Anonymous Coward | about 2 years ago | (#41960793)

It ships with Compute Node Linux, which is a cut down (lower overhead) version of SLES. It supports several schedulers, but ORNL typically uses Altair PBS on the big systems (http://www.cray.com/Products/XK/Software.aspx [cray.com] ). ORNL provides a large number of compilers and libraries that users can use in the form of 'modules' (http://www.olcf.ornl.gov/support/user-guides/titan-user-guide/ [ornl.gov] ). And in terms of scheduling/partitioning, the user just requests a specific number of nodes when they submit a job, and they get those nodes to themselves for the allotted time. It's pretty low-impact on the compute nodes, and less exciting than you might think. They don't put much emphasis on the software when reporting on these machines, because it's stripped down as much as possible to allow the user applications to run at peak performance.

I would love.... (0)

Anonymous Coward | about 2 years ago | (#41960335)

to get into Titan with a magnetron....

My Macbook (2)

gnasher719 (869701) | about 2 years ago | (#41961639)

would have made it onto the June 2000 supercomputer list, and would have made 2nd place in the June 1993 list, both with the CPU alone. At least if Dongarra wrote the Linpack code :-)

2.3 GHz, 4 cores, 256 bit vector registers, one add + one multiply per cycle throughput = 73.6 GFlop/s theoretical limit; with hyperthreading one should get quite close to that limit. Top 500 in June 2000 was only 44 GFlop/s, 2nd place in June 1993 was 30 GFlop/s.

Re:My video card (0)

Anonymous Coward | about 2 years ago | (#41962049)

Meh. I have an $80 video card that does around 700 GFlop/s, and they sell cards that do 5600 GFlop/s for $1000 MSRP (GTX 690). The supercomputer in question does 17,000,000 GFlop/s. That's equivalent to 3,000 GTX 690's, or about 24,000 of my video card, or about 230,000 of your laptop.

p.s. If "Moore's Law" holds up, they might be selling ~$1000 video cards as fast as the supercomputer in 2029, or $80 video cards that fast in 2033, or laptops that fast in 2038.

Hmmm (1)

slick7 (1703596) | about 2 years ago | (#41962163)

Are we in a race to see just how fast we develop AI that will kick our asses? Skynet or the Forbin Project seem to be the goal.

Based on AMD Opterons (0)

Anonymous Coward | about 2 years ago | (#41962331)

Not only it is based on AMD Opterons, but on Opteron 6274, the Bulldozer architecture, which is bashed by every (desktop) review and here on Slashdot. It is funny that a "completely useless" processor powers the fastest computer in the world.

To be fair, only Phoronix does real server benchmarks, and even they have a different objective, not processor benchmarking. And to a large extent every benchmark is useless, you always have to benchmark your own application to get results which can be taken seriously.

Hardware vs Software limits (0)

Anonymous Coward | about 2 years ago | (#41964899)

I think we are reaching the point where we are not limited by hardware as much as software. It seems like all these ultra-fast supercomputers can do is fairly simple calculations but at a massive scale. I bet we could do a lot more with the hardware we have now if we developed more subtle and complex software similar to how brains function.

Hard to discover which OS the computers use. (2, Informative)

Anonymous Coward | about 2 years ago | (#41965419)

The obvious question "which are running Linux" is almost impossible to answer. Are they sponsored by Microsoft? However, you can use the "sublist" feature to make a list of the first 500 computers and limit to "Linux" as operating system. The list contains 469 entries, and the first number that is missing from this list is "38".

Huh.

So while it will refuse to actually show the OS, the sublisting feature makes it able to wrestle it out indirectly.

Re:Hard to discover which OS the computers use. (0)

Anonymous Coward | about 2 years ago | (#41980053)

You can download the Excel version of the list and all of those details are readily available...

Distributed Processing? (0)

Anonymous Coward | about 2 years ago | (#41966309)

Will Colussus and Guardian be measured separately or as a single device?

Check for New Comments
Slashdot Login

Need an Account?

Forgot your password?