World's fastest:IBM's Roadrunner supercomputer breaks petaflop barrier. Cell+Opteron

[edesignuk]

When you're looking to set a record this is how you do it. Not only has IBM's Roadrunner supercomputer come on-line, it's now the world's fastest -- twice as fast as the old BluGene/L champ -- and churning through 1.026 quadrillion calculations per second. The $133 million supercomputer achieved the milestone with the help of 12,960 "improved" Cell processors (yes, like those powering your PS3) and a smaller number of AMD Opteron processors -- 116,640 processor cores in total. Unfortunately (or fortunately depending upon your perspective), Roadrunner is for military use only so you'll have to solve the traveling salesman problem on your own time. While not quite into Exaflop territory, we're definitely on the way.

Engadget.

 

[Abstract]

Should have gone with Intel Inside™, and put Intel stickers all over the machine.

 

[David G.]

116640 processor cores

What! That's it? My homebuilt beowulf cluster has more than that.

 

[Counterfit]

Cell is somewhat sorta-kinda based on PPC isn't it? Interesting that they chose two fairly different architectures for one unit.

 

[QuantumLo0p]

Crazy! I never see one of these things and walk away unimpressed.

A few years back I had an opportunity to tour the SGI super computing lab on the WAM!NET campus during a regional Unigraphics user group conference. If I remember correctly, SGI's biggest iron at the time, had 32Gb/sec input/output.

I wonder about the complete specs of IBM's newest beast. Imagine where we will be in ten years. Yikes. Holodeck, here we come.

 

[Consultant]

but it STILL doesn't hold all the p0rn on the internet. =p

Maybe in 100 years iPhones will be able to compute the meaning of life...

 

[BBC B 32k]

could they get one of these inside a powerbook?

 

[dmw007]

Cell is somewhat sorta-kinda based on PPC isn't it? Interesting that they chose two fairly different architectures for one unit.

My thoughts exactly, a Supercomputer using both PowerPC and x86? Quite the machine though!

 

[Counterfit]

but it STILL doesn't hold all the p0rn on the internet. =p

I bet if it had 1GB per core, it could come close.

 

[edesignuk]

I bet if it had 1GB per core, it could come close.

I think you grossly underestimate the pornage of the interwebs, young one

 

[velocityg4]

I wonder what kind of frame rates that gets in Quake?

 

[themadchemist]

could they get one of these inside a powerbook?

They tried. But we all know that IBM can't get its processors to run cool enough, and that's why we don't have 100,000 Cell processors inside a PowerBook. I mean, I'm glad Apple switched from IBM--if you can't even do that, what's the point of even existing?

 

[PlaceofDis]

thats just insanity. i hope its put to good use. like finding the answer to the universe, life, and everything.

 

[themadchemist]

For some perspective...

Processing power of the human brain.

By comparison, the human brain uses about 20 watts or about 100 watts to power the entire body. And the average household light bulb uses 40-100 watts.

Furthermore, there is no real consensus regarding the computational power of the human brain. Some estimates suggest that it is capable of 1017 FLOPS, or 100 petaflops.

According to Hans Moravec of Carnegie Mellon University (whom used the processing speed of the retina), estimates that the brain operates at 100 million MIPS (or is it 100 teraflops?). Nick Bostrom of Yale estimates it to be 1,000 times higher than Moravec’s number (or 100 petaflops). [Note: one must not conflate specialized hardware designed to defeat "grandmaster" chess players with the wide ranging cognitive abilities human brains are capable of].

Estimates range from 100 teraflops to 100 petaflops (so 100 to 10,000 times the processing power of this computer). However, the complexity of human circuits and the diversity of the circuit modalities (the various types of transmitters, morphologies of the neurons, levels of insulation, spatial oragnization, etc.) suggest to me that even if we are getting kind of close to the process/second abilities of the human brain, we are far, far away from its actual functional power.

Oh yeah, and it does all this with 20 watts.

It'll be interesting to see if, beyond extending to the nth power the computational muscle of our computers, we find ways to mimic the brain a little more...Perhaps by increasing the diversity of the processes on these chips in ways that resemble neural circuits, we might be able to extract more complex functions with less of an increase in power consumption.

 

[PlaceofDis]

For some perspective...

Processing power of the human brain.



Estimates range from 100 teraflops to 100 petaflops (so 100 to 10,000 times the processing power of this computer). However, the complexity of human circuits and the diversity of the circuit modalities (the various types of transmitters, morphologies of the neurons, levels of insulation, spatial oragnization, etc.) suggest to me that even if we are getting kind of close to the process/second abilities of the human brain, we are far, far away from its actual functional power.

Oh yeah, and it does all this with 20 watts.

It'll be interesting to see if, beyond extending to the nth power the computational muscle of our computers, we find ways to mimic the brain a little more...Perhaps by increasing the diversity of the processes on these chips in ways that resemble neural circuits, we might be able to extract more complex functions with a less of an increase in power consumption.

thanks for that. thats amazing what our brains are capable of. but we don't want machines to mimic the human brain. for then it will be the beginning of the robot age, where we must submit to our mechanical overlords.

is there any estimated processing power between different species? or are all living creatures pretty much processing at the same speed/power etc. that makes me wonder about this 'consciousness' thing....

 

[Mousse]

thats just insanity. i hope its put to good use. like finding the answer to the universe, life, and everything.

The article said it's ONLY for military use. I guess that means they'll use it for nuclear simulations or balancing the budget or playing chess.

Yeah, supercomputers like playing "a nice game of chess."

 

[iJohnHenry]

Will no one photochop a Mac sticker on that beast??

 

[dmw007]

For some perspective...

Processing power of the human brain.



Estimates range from 100 teraflops to 100 petaflops (so 100 to 10,000 times the processing power of this computer). However, the complexity of human circuits and the diversity of the circuit modalities (the various types of transmitters, morphologies of the neurons, levels of insulation, spatial oragnization, etc.) suggest to me that even if we are getting kind of close to the process/second abilities of the human brain, we are far, far away from its actual functional power.

Oh yeah, and it does all this with 20 watts.

It'll be interesting to see if, beyond extending to the nth power the computational muscle of our computers, we find ways to mimic the brain a little more...Perhaps by increasing the diversity of the processes on these chips in ways that resemble neural circuits, we might be able to extract more complex functions with less of an increase in power consumption.

Very interesting read themadchemist, thank you for sharing that information with us.

The human brain is extremely energy efficient if it is capable of that amount of processing power and only consuming a mere 20 watts of energy!

 

[Consultant]

Will no one photochop a Mac sticker on that beast??

I found this photo in the elevator! Somebody must have dropped it.

It's inside an envelope labeled "Secret Photo of the new PowerBook G5 (Code Name: Meep Meep)".

 

[themadchemist]

thanks for that. thats amazing what our brains are capable of. but we don't want machines to mimic the human brain. for then it will be the beginning of the robot age, where we must submit to our mechanical overlords.

is there any estimated processing power between different species? or are all living creatures pretty much processing at the same speed/power etc. that makes me wonder about this 'consciousness' thing....

I imagine that there are considerable differences, though I don't know whether this has been explored (all of this is a bit outside my field). But the difficulty in comparison probably relates to the fact that in addition to differences in sheer processing capability, there are probably tough-to-quantify differences in the number of interconnections, the kinds of neurons, and the circuits that exist for special types of processing. In a way, you might liken this (I think, now we're going way outside my field) to the instruction sets on a processor that help it gain efficiency in particular functions. Scaling up doesn't uniformly increase performance, but can do so in targeted ways. And increased executions/second don't necessarily translate linearly into a given desired output.

 

[dmw007]

I found this photo in the elevator! Somebody must have dropped it.

It's inside an envelope labeled "Secret Photo of the new PowerBook G5 (Code Name: Meep Meep)".

Great find Consultant, I always knew that it was only a matter of time for the PowerBook G5 to make its debut!