Contrary to popular belief, the computers we don't know about aren't really bigger/faster than the ones we do. There are a lot of them, we don't even know how many or how big exactly, but I work with a whole bunch of people who have direct knowledge and nothing they've ever said has even hinted at the secret existence of machines faster than Roadrunner. It would be impossible to hide such a thing from the entire HPC community (many of whom would need to be involved in its design and deployment) so in this case absence of evidence really does translate into evidence of absence.
I don't believe that Roadrunner is really the fastest computer either, though, for a completely different reason. If you put both (XT4 and XT5) parts of Jaguar at ORNL together, which is not only technically feasible but has been done, the sum is faster than Roadrunner. Since LINPACK is embarrassingly parallel anyway, that would be an entirely valid comparison, and it could also be argued that the last-minute upgrade to Roadrunner came too late to qualify for the top spot at SC08. On many real applications other than those specifically written for it, hellish-to-program Roadrunner might well struggle to get into the top five. BTW, Jaguar is not scheduled to go "behind the curtain" either. It's dedicated to open science and likely to remain that way.
In Roadrunner's defense, I do have to make one other point. It is extremely power-efficient compared to other top-tier systems, even to its Blue Gene cousins, and I think that's quite an achievement in its own right.
Oh, I forgot to mention that several systems are already being built or upgraded that will eclipse these systems. Baker and Intrepid are the two that come to mind in the near term, Blue Waters further out. Google for them if you want to have your mind blown.
over 16,000 AMD Opteron processor cores and over 16,000 Cell B.E. processors...will occupy only 12,000 square feet of floor space, or approximately the size of three basketball courts...
I have heard it said, from someone I used to work for who claimed to have been in on such things, that IBM's laboratory stuff was very far ahead of everyone else in the 1980s because of work they did with the NSA, and that they even had copper-wired microprocessors long before they announced it. I can believe this might actually have been true then, but considering nothing of interest has happened with the technology since Intel demonstrated a 10GHz ALU about a decade ago, I find it hard to believe this is still the case.
Besides, clustering makes it unnecessary anyway -- if you need more power, throw another rack of blade servers in the room.
The fastest publicly announced computer ... I am fairly certain that this is not the fastest computer, but it is relatively close.
Contrary to popular belief, the computers we don't know about aren't really bigger/faster than the ones we do. There are a lot of them, we don't even know how many or how big exactly, but I work with a whole bunch of people who have direct knowledge and nothing they've ever said has even hinted at the secret existence of machines faster than Roadrunner. It would be impossible to hide such a thing from the entire HPC community (many of whom would need to be involved in its design and deployment) so in this case absence of evidence really does translate into evidence of absence.
I don't believe that Roadrunner is really the fastest computer either, though, for a completely different reason. If you put both (XT4 and XT5) parts of Jaguar at ORNL together, which is not only technically feasible but has been done, the sum is faster than Roadrunner. Since LINPACK is embarrassingly parallel anyway, that would be an entirely valid comparison, and it could also be argued that the last-minute upgrade to Roadrunner came too late to qualify for the top spot at SC08. On many real applications other than those specifically written for it, hellish-to-program Roadrunner might well struggle to get into the top five. BTW, Jaguar is not scheduled to go "behind the curtain" either. It's dedicated to open science and likely to remain that way.
In Roadrunner's defense, I do have to make one other point. It is extremely power-efficient compared to other top-tier systems, even to its Blue Gene cousins, and I think that's quite an achievement in its own right.
Oh, I forgot to mention that several systems are already being built or upgraded that will eclipse these systems. Baker and Intrepid are the two that come to mind in the near term, Blue Waters further out. Google for them if you want to have your mind blown.
The press release with some numbers
I have heard it said, from someone I used to work for who claimed to have been in on such things, that IBM's laboratory stuff was very far ahead of everyone else in the 1980s because of work they did with the NSA, and that they even had copper-wired microprocessors long before they announced it. I can believe this might actually have been true then, but considering nothing of interest has happened with the technology since Intel demonstrated a 10GHz ALU about a decade ago, I find it hard to believe this is still the case.
Besides, clustering makes it unnecessary anyway -- if you need more power, throw another rack of blade servers in the room.