Talk:Roadrunner (supercomputer)

From Wikipedia, the free encyclopedia
Jump to navigation Jump to search
WikiProject Computing / Hardware (Rated C-class, Low-importance)
WikiProject iconThis article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.
C-Class article C  This article has been rated as C-Class on the project's quality scale.
 Low  This article has been rated as Low-importance on the project's importance scale.
Taskforce icon
This article is supported by Computer hardware task force (marked as Low-importance).

Update NEEDED![edit]

Introduction totally wrong and clearly outdated!
Nowadays, roadrunner is 19th on the last list and I couldn't find it anywhere (quick search) on the greentop500 list..

I updated it and removed date sensitive wording. -- Henriok (talk) 14:20, 7 September 2012 (UTC)


"Purpose: used for predicting outcomes of nuclear war" ???? —Preceding unsigned comment added by (talk) 21:51, 31 July 2008 (UTC)

What's been said is that they are simulating the decay of the nuclear stockpile, and that they are simulating nuclear blasts (since they are forbidden to do live tests). I don't know If they are doing simulations of war scenarios. I think the edit was a prank, perhaps a reference to the Joshua computer in the movie Wargames, whos purpose was to simulate and act upon thermonuclear war. But, I think it's pretty plausible that they are doing such experiments too. I would. -- Henriok (talk) 22:46, 31 July 2008 (UTC)


I FAIL! :)

Actually is there any news on how much it costs? Thanks --Drmike (talk) 14:27, 10 June 2008 (UTC)

It cost USD 100 million. Anonymous101 (talk) 16:48, 10 June 2008 (UTC)

Wrong focus[edit]

The system operations section is probably just great for systems operations experts. The trouble is, this is an encyclapedia for the general reader.. --Philopedia (talk) 15:08, 10 June 2008 (UTC)

Perhaps someone could also include a simplified system description so that non-techy people could understand the computer along with the techy types? (talk) 16:53, 10 June 2008 (UTC)

it's a beautiful article, but yeah, I'd like lots more layman stuff - for example, an explanation of what a 'blade' is Adambrowne666 (talk) 18:22, 10 June 2008 (UTC)

I think a more simplified answer is it has the computing power of 100,000 of today's most powerful laptops. I found this on Can anyone confirm this as a reliable source?RobSoko315 (talk) 21:44, 10 June 2008 (UTC)
It may be fine for high tech folks, but it's simply incomprehensible to anyone else. "Wrong focus" is a good phrase for it. The article also lacks citations in the text. And, finally, the whole text sounds like an IBM advertisement. Unacceptable by several different Wiki standards, I'd say. Timothy Perper (talk) 21:49, 10 June 2008 (UTC)
TGDaily also have a reference of the 100,000 laptops.[1]. I think to a layman the concept of a computer that is capable of doing the same work as 100,000 laptops is perfectly acceptable, in the same way a layman understands that his 200bhp car is twice as powerful as one that is 100bhp. Just my two cents. -- (talk) 10:51, 13 June 2008 (UTC)
It's not a good comparison for a couple of reasons. Laptops today vary a great deal in performance. Consider the ASUS Eee PC to a Apple MacBook Pro, which is at least five times as fast. And, this is today's laptops. One year from now, who's going to keep this comparison up-to-date? A person interessted in the details of how fast this computer is, can examine the article about flops, the article about Top500 (coming up soon, Roadrunner is not on that list yet), or just imagine a computer costing 133 million dollars which draw as much power as a shopping mall, is 50x100 m large and cosisting of nearly 20000 processors of the lastest crop. -- Henriok (talk) 23:48, 14 June 2008 (UTC)

Does it come with a copy of the Internet? Or else I won't buy it! —Preceding unsigned comment added by (talk) 13:14, 9 November 2009 (UTC)

what is "next-generation" except for a buzzword?[edit]

Please define. The article on supercomputers does not define the term either. -- (talk) 15:56, 10 June 2008 (UTC)

I would assume that it would mean the movement from terascale to petascale to (future) exascale. It could also mean the concept of using a hybrid GPGPU/CPU system. I agree that the phrase is vague in the context. fintler (talk) 16:06, 4 January 2011 (UTC)

computing power[edit]

Would this computer be able to predict what my girlfriend's mood is going to be like today? Or have we not reached that level of computing power yet? (talk) 19:06, 10 June 2008 (UTC)

The answer is already underway, progress is being made, and a solution is expected shortly.WurmWoodeT 20:01, 10 June 2008 (UTC)

Major piece of info missing[edit]

The article says that it was constructed for the DOE, but it doesn't say what the DOE needs it for. It might be important to know what the world's most powerful computer is being used for... VanTucky 20:04, 10 June 2008 (UTC)

DOE refers to the "Department Of Energy" united states... —Preceding unsigned comment added by (talk) 14:49, 11 June 2008 (UTC)

How does this compare to a normal computer?[edit]

I've heard from conversation that the processing output of Roadrunner for one day is equal to 6 billion normal computers working for 40 years. Can anyone confirm this and find a reliable source to cite it? RobSoko315 (talk) 21:32, 10 June 2008 (UTC)

Off, by quite a few orders of magnitude. A normal desktop has a few gigaflops of processing power; a few dozen more if you count the graphics cards' raw performance capabilities. That means this computer is on the order of 100,000 times faster than a desktop, not on the order of trillions of times faster. (talk) 23:51, 10 June 2008 (UTC)
However it would be slower than most desktop computers for the majority of tasks, as its parrallel where it gets it speed. Sequentially its not any faster than a desktop and no general user programs will run faster on it. The only thing it is good at is parrallel distributed processing.-- (talk) 03:06, 11 June 2008 (UTC)
Also note how many processor cores were used in its construction. If the Intel processors were single-core (which this article doesn't say otherwise) there are 26,784 processing cores running to generate this 1.026 petaflop operation. That translates to roughly 38.3 gigaflops per core. If Sony is to be believed, this will take around 3850 (maybe a little less) PS3s for one second. (talk) 03:36, 12 June 2008 (UTC)
There are no processors from Intel in Roadrunner, they are from IBM (the PowerXCell) and from AMB (Opteron). This is a hybrid system with different parts doing different kinds of computations. The Opterons are doing one thing (operating system, networking, some computation), and the Cells are doing another (heahy lifting, numerical algorithms, floating point computations). When IBM is spec:ing the performance for the Cell processor in the PS3, they are talking about single precision floating point operations. Linpak (the benchmark used by Top500) does double precision, which the processor in SP3 suck at. PowerXCell is at least 5 times as fast in this respect. So.. 20.000 PS3s would be a more accurate estimation, if you ignore the fact that the interconnect would be Ethernet, not Infiniband. Looking at the supercomputers on the Top500 list, none on the top 50 uses Ethernet, and it would probably degrade overall performance for the complete system by 30-40% or so. So.. perhaps 25.000 PS3s would be the final estimation of how many it would take to make a similarly powerful system, if it would work at all. -- Henriok (talk) 07:34, 12 June 2008 (UTC)
Man, I'd bet you could get some killer PvP action going on that machine. I'd probably just end up using it to play Age of Conan. —Preceding unsigned comment added by (talk) 22:17, 12 June 2008 (UTC)

If this system is equivalent to 25000 PS3, then where is the cost justification? 25000 PS3s = $10 million (at $400/PS3), while this machine cost a 100 million... that's 1000% more expensive... Which gives two possible explainations: 25000 PS3 estimate is wrong, or people who made this machine are idiots (talk) 01:00, 17 September 2008 (UTC)

There's more to this than hardware. The cost inludes desining the system, designing software models, operating systems, staffing, housing, cooling, storage, networking. Roadrunner has been under development for a couple of years employing quite a few highly paid engineers. And I see that I was comparing the teoretical performance of a PS3 to the actual performance of Roadrunner. The theoretical peak of Roadrunner is about three times as high, so I should have made the estimation more like 75000 PS3s. And I really don't think it's possible to build a meaningful supercomuter of this magnitude using just PS3s so the estimation might bee moot to begin with. I think you can do an equally invalid comparison by chucking in a couple of thousand GPUs and just multiplying the cost of the graphics card to reach the desired performance. -- Henriok (talk) 07:45, 17 September 2008 (UTC)


How does a computer have "almost 6,912 AMD Opteron dual core processors and almost 12,960 IBM PowerXCell[5] 8i CPUs." It either has that many or it doesn't... —Preceding unsigned comment added by (talk) 21:38, 10 June 2008 (UTC)

I noticed that too. It's 3^3 *2^8, which perhaps has something to do with the fact that Roadrunner apparently is organized in 3s at various levels? (talk) 23:22, 10 June 2008 (UTC)

Agreed. This sentence must be revised. "almost 6,912 ... processors" is ridiculous. It should read 6,912 or almost 7,000. --Dweller (talk) 06:57, 11 June 2008 (UTC)
Well it is clearly possible for a computer to have almost 6,912 and 12,960 processors (for example a computer with 6911 and 12959 processors). However I agree with Dweller that the statement is pretty silly. Nil Einne (talk) 11:15, 11 June 2008 (UTC)
Further, shouldn't it read 12,960 IBM PowerXCell 8i CPUs and 6,912 AMD Opteron dual core processors? (Higher number before smaller number?) -- (talk) 04:08, 12 June 2008 (UTC)
The article makes clear that the PowerXCells are being used attached to the Opterons, which are the ones that are interconnecting with each other and managing "housekeeping". It takes a LOT of coordination to have over 10,000 cells processing a data set simultaneously. So the each Opteron is the "master" of a cell, and spends its time coordinating with the other cells. The "slave" processors, the Power chips, are the ones crunching the numbers. Counting the masters before the slaves is entirely reasonable. --Eliyahu S Talk 13:03, 1 July 2008 (UTC)
In a super-cluster like this, with tens of thousands of cards, at almost any given moment at least one, and probably several, of the processor cards will be "hung", "on the fritz", or in some other way not working. So while there might be 6,912 Opteron cores, with 6,480 of them each connected with 2 PowerXCells, (totaling 12,960,) physically plugged in, ALMOST that many of them will be working at any point in time. --Eliyahu S Talk 13:03, 1 July 2008 (UTC)

Problems here[edit]

I'm adding a second comment to go with the one I put above. This article lacks definitions of technical terms, it sounds like an advertisement, it lacks citations to specific claims, and it's completely unclear to anyone unfamiliar with high-speed digital compiter hardware and software. For example, what is a "petaflop"? It sounds vaguely obscene and no, it is NOT obvious that a "petaflop" is a measure of computational speed. I'd flag this article with all kinds of comments, like "needs citations," "original research," and related criticisms, but maybe this comment will stimulate someone to rewrite the article to Wiki standards. It sure needs it. Timothy Perper (talk) 21:59, 10 June 2008 (UTC)

We have an article on FLOPs. It is already wiki linked. While I agree the article could so with some work, I don't think anymore is needed in that particular area. It is not necessary to explain what petaflop means any more then it is necessary to explain what a watt is in an article on a power plant. People unfamiliar with what amounts to basic terminology are free to check out the wikilinked articles, that's the whole point with wikilinks Nil Einne (talk) 11:16, 11 June 2008 (UTC)

Can't say I agree. And I noticed that someone else added an overall comment about clarity and jargon. Thanks.

If one actually clicks on the petaflops wikilink, one gets to "flops." The only time "petaflops" appears is in the text box at the top, where it says that a petaflop is 1015 flops. So? What's so special about petaflops? Actually, I was using petaflops as an example of a problem that runs through the whole article, but I guess I wasn't clear enough. OK, here it is again,

For example, what is an Opteron core? Or a aPowerXCell 8i CPU? Or an eLS21 Opteron blade, an expansion blade, or a QS22 Cell blade? What us an Infiniband 4x DDR adapter? What is a 288-port Voltaire ISR2012 Infiniband switch? Why are any of these even remotely interesting or significant?

Let me say as clearly as I can that all this sounds like somebody copied a section of a technical manual directly to Wikipedia. Or, worse, they copied an advertisement or publicity blurb from IBM.

And yes, it is necessary to explain what a "watt" is -- it's a unit of electrical power. Now, what's a petaflop? The answer is NOT that it's 1015 flops. Why not? Let's say I'm writing a wiki article about gnushes, and mention that the RTPX gnush contains 1024 polygnushes. Someone asks me what a polygnush is, and I tell them that it's a subordinated array of from 24 to 512 gnushes. That is no help, no help AT ALL.

In brief, if you write a technical article for the general public -- and Wikipedia has to be written for the general public, because that's who reads it -- you have to explain technical terms. That is not optional: you can't fob it off on another, equally opaque article.

Timothy Perper (talk) 15:48, 11 June 2008 (UTC)

Not all articles has to or can be written towards a lay person. This is an article about the worlds fastest computer, it's bound to include "jargon" and technical stuff, since this is the bleeding edge of computing. Most of the terms above isn't anything that's complicated for a person who is interessted in processing technology or supercomputing, or even high-end desktop computing. Terms are explained on the respective page. What's wrong with this page might be that it does not link enough to other articles. What a flops is is sufficiently explained on its page, and if it's not, it's an issue for that article, not this. It is however clearly explained on that page. ALL articles require some level of prior knowledge, and that level differ from article to article. Basic education is a good start. One must expect the general public to be able to use the search field, and be able to click on provided links. Explainging technical terms in exhausting detail on every page that uses them is just moronic. If someone want to know more about supercomputing, there's plenty of resources on Wikipeida for that, the same goes to how a server is built, what components go into a computer, what dows what, and why, and how different kinds of computer networks work. This article is not the place for those topics. Highly specialized topics require that the casual reader does his homework. For the really casual reader the sentance "Currently the world's fastest computer" should suffice, and there'sno need do define either "currently", "world", "fast" nor "computer". At least not here. -- Henriok (talk) 22:36, 11 June 2008 (UTC)
I don't agree. Words like "fast" need to be defined, because they're important in this field. And the people writing this article are NOT talking only to themselves, but everyeone who sees the home page of Wiki and clicks on the Roadrunner link. My questions remain:
"For example, what is an Opteron core? Or a aPowerXCell 8i CPU? Or an eLS21 Opteron blade, an expansion blade, or a QS22 Cell blade? What us an Infiniband 4x DDR adapter? What is a 288-port Voltaire ISR2012 Infiniband switch? Why are any of these even remotely interesting or significant?"
Just click on the links provided in the article, and the links that are provided on those pages and so on and so on. This is the way World Wide Web works, and Wikipedia. In that manner, you'll learn stuff you wouldn't otherwise know. You cannot collect all the neccessary information for one article under that article. That is what is moronic, and that was what I said. If you still find it a good idea, I'm sorry but Wikipedia can't help you then. Here is some further reading: Opteron, Multi-core, Cell (microprocessor), Blade server, IBM BladeCenter, Infiniband, Computer port (hardware), Network switch, Double data rate. I believe that all the links is provided in the article, in a manner that satisfies Wikipedias manner of style. "Voltaire" is a company (without a Wikipedia article) and "ISR2012" is a product they make. Defining what's interessting and significant is a personal judgement, and Wikipedia should try to keep a neutral point of view so that shoud be left out. Most of the information you require is behind the links however. Please do click them. -- Henriok (talk) 20:25, 12 June 2008 (UTC)
If you want to write jargon, post all this to a blog or a chat group for computer groupies. If you want to write on Wikipedia, it's necessary, IMO, to write for people who can snow *you* with jargon from their own fields, but who go out of their way to try to explain what the jargon means and why it's important. I do not believe that Wikipedia entries can hide behind a "I am expert, and you're a NOBODY" way of thinking. That is NOT what Wikipedia is all about, at least not to me.
Please show me an article that lives up to your standards. Let's take todays featured article, the durian, an article that should live up o Wikipedia's higest standards. In the first paragraph, I found these missing definitions (in order): "tree", "large", "family", "widely known", "revererence", "formidable", "thorn", "growth", "30", "centimetre", "inch", "long", "diameter", "typicallity", "weight", "kilogram", "pound", "shape", "range", "oblong", "round", "color", "green", "brown", "flesh", "pale", "yellow", "red" and "dependancy". I'm not into the fields of biology, physics, mathematics or physical appearances, but there seems to be a lot of jargon here.. Please explain why all this is significant. Of course I'm going off on a tangent here just to make a point, but nearly as you can see, most of these terms actually do have articles of their own on Wikipedia so they are important enough to write articles about. The rest seems largely to be breaches of the NPOV rule.I think you'll find articles like Trigonometric functions to be just unbearable with jargon just all over the place and it's not even complicated maths. -- Henriok (talk) 20:25, 12 June 2008 (UTC)
And maybe you should read Wiki's advice about being civil before calling people or their ideas "moronic."
I didn't call anyone moronic, but the idea (and I quote myself in verbatim) "Explainging technical terms in exhausting detail on every page that uses them" still is I'm afraid. Wikipedia should not be a single page, infinitely long, explainging every aspect in and of the universe. I can distiguish between and separate a person and the ideas they have. If I did hurt the feelings of an idea, I do apologize. -- Henriok (talk) 20:25, 12 June 2008 (UTC)
Timothy Perper (talk) 14:41, 12 June 2008 (UTC)
You should read Wiki's advice about being civil -- you really should. So far, some three or four people have pointed to serious issues and problems with this article. I'm not alone in feeling that this article has genuine problems with incomprehensibility. Terms like "triblade" and "Opteron" are not common words of English like "tree" or "formidable." They're jargon in a specialized field. They need explanations. The trigonometry article is (IMO) irrelevant; we're talking about this article, not trigonometry or all of Wikipedia. That's called "doing one thing at a time." May I suggest, politely but firmly, that perhaps you might devote your time to supplying those explanations rather than being sarcastic at my expense? Timothy Perper (talk) 21:40, 12 June 2008 (UTC)
Arguing on the Internet is like competing in the paralympics.. I did some work instead. How about my additions? Triblade is already explained in the article and Opteron realy, really is a common component in computing. You should get used to the fact that all articles isn't and can't be written to satisfy a reader withour any prior competence in the area. If one were to stuble upon a topic that seems to be buszzing with jargong what one isn't familiar with, one should consider that the error is not that the topic isn't explained sufficiently but perhaps one just isn't educated enough. Ranting on, blaming others for one's own ignorance isn't really the best way to go -- Henriok (talk) 23:05, 12 June 2008 (UTC)

This is going nowhere. You're being uncivil, aka offensive. Do not blame the reader when what is written is unclear. And this article is not clear. But I'm not going to discuss it further with you. Timothy Perper (talk) 04:37, 13 June 2008 (UTC)

As I see it (maybe I'm wrong), Henriok's incivility consisted entirely of calling one thing (namely, explaining all technical terms) moronic. Your analogy is that if someone mentions RTPX gnushes and explains that those are a certain number of polygnushes, which are a certain number of gnushes, you'll still have no idea what he's talking about. This article mentions petaFLOPS and links to the article on FLOPS; that article goes on to state that a petaFLOPS is a certain number of FLOPS, which it does define as "floating point calculations per second", with links to "floating point" and "calculations". Therefore, I don't agree that petsFLOPS is not sufficiently defined.
As another analogy, an article on, say, Fourier transforms is bound to mention integrals if it wants to go into any (well, much) depth, but it would be a bad idea to try to explain what an integral is on a page about Fourier transforms: integrals are difficult to explain, and though integrals are a technical term, they are one of the two most important concepts in calculus, so any person reading the definition of a Fourier transform is likely to know what an integral is; if they don't, they can read about integrals on Wikipedia's page on integrals. -- (talk) 19:02, 28 June 2008 (UTC) (Ihope127 (talk))
You are correct. –– Henriok (talk) 19:37, 28 June 2008 (UTC)

Power consumption?[edit]

The one datum I was looking for, viz. how much power this thing consumes, wasn't here. It might also be nice to know how it's cooled, as all that power has to be dissipated somewhere. Shalom S. (talk) 21:48, 11 June 2008 (UTC)

3.9 MW. Just added that fact. -- Henriok (talk) 23:08, 12 June 2008 (UTC)
It's 2.345 MW power. It's what they measured when they benchmarked the computer with LINPACK for the Top500 entry. 3.9 is probably a theoretical peak. -- Henriok (talk) 13:35, 18 June 2008 (UTC)
It's possible that it's both at once. The cluster's calculated power consumption of 2.345 MW would be what the blades draw, while the 3.9 MW might be the measured load of the entire facility, including the extra power wasted by hundreds (wow! scary!) of cabinets, each with its own power bucket and blowers at the bottom, together with the huge air moving and cooling facilities, and tertiary electrical costs for a center like lighting, consoles, the security and fire safety systems, etc. So the DOE is paying the bill on a 3.9 MW feed (which is a medium-large industrial connection) and out of that the CPU's themselves are using 60% of the total. --Eliyahu S Talk 13:16, 1 July 2008 (UTC)
The 3.9 MW figure was gathered from a presentation that was held before Roadrunner was fully operational, so this is most likely a theoretical figure. 2.345 MW on the other hand was measured while running the benchmark test. It's a real world figure, on high load. has a pretty good description of what they did and did not measure when they got this figure. Check it out! -- Henriok (talk) 14:11, 1 July 2008 (UTC)
Thanks for the link. From what I read there, it looks like they did NOT include any of the "ancillary" power consumption possibilities that I listed. 2.345 MW is just the processors themselves, while running the benchmark. For all I know, the 3.9 MW figure might indeed be the projected maximum for the entire facility. I'd be curious to know what their TOTAl consumption was. --Eliyahu S Talk 15:19, 3 July 2008 (UTC)


The "System Overview" section opens with a description of a "Triblade", with no context or explanation given as to why how triblade relates to the system. An introductory sentence is needed explaining that the system is built up of fundamental units called "triblades" (if indeed that is the case, which I can't tell from the current description). Alternatively, the system can be described in the reverse order that it is currently, which is in bottom-up fashion, from lowest-level building block to the complete system. It could be described instead in a top-down fashion, describing its major components, then the subcomponents, etc., down to the lowest fundamental computational unit. This might make the description more comprehensible. — Loadmaster (talk) 15:17, 12 June 2008 (UTC)

You bring up a good point. When I first read the article, I myself had to backtrack, or rather move forward to see why I needed to know about a Triblade. Comprehensibility of the article will be improved by describing it in a top-down fashion, as readers are interested in the main thing itself, and then will further read on if they are interested or want to know more. Right now, readers need to cut through the connected units and the description of the triblade to read about the Roadrunner itself. Further, it's easier to break things down, instead of build things up especially with such technical jargon. However, if the article won't work in a top down fashion, there should at least put an introductory sentence before the description of the Triblade. (And, on further reading of the article, it looks like it is built from triblades.) ProjectTux (talk) 22:58, 12 June 2008 (UTC)

Yes, I agree. I ran across this article on the Wiki home page and it looked interesting, so I clicked on it. Computers are not my field, though I recall fondly many years ago indeed soldering together an 8080 with other bits and pieces to make this contraption that flashed lights when it added 1 + 1 and got 2. So I am not computer unfriendly, but I'm not a technogeek either, just interested. And yes, you are absolutely right: I'd like a general description of Roadrunner, with a top-down description of its architecture and innovations. In one place, the article says Roadrunner has been built (opening paragraph), but later on, someone says it's unfinished. Which is it? I think the article also needs a discussion of what Roadrunner will be used for, if only to explain why the DOE or LANL or the military needs the device. Once again, the whole point of Wiki is that it is NOT -- not -- a specialist's bulletin board or chat room; it is a general encyclopedia written for people in the general public who do not know as much as the author(s) about the topic. None of us is expert in everything, and we all need explanations of things we don't know. Roadrunner sounds like an impressive little toy -- and the article needs a lot of work to tell us how it operates and what it does, in brief, why it was worth 122 million dollars of the taxpayer's money.
Most of the money goes into housing, staffing, cooling and programming. The actual hardware isn't even close to the largest part of the cost. There are probably 20 people or more (highly payed scientists and engineers) working full time on making this baby purr. That's not cheap, and they have already worked for 2 years on it. The power bill is substantial as you can imagine. -- Henriok (talk) 13:13, 13 June 2008 (UTC)
About power consumption: 3.9 megawatts. MW is wikilinked, and here's what the link says about megawatts.
Many things can sustain the transfer or consumption of energy on this scale; some of these events or entities include: lightning strikes, large electric motors, naval craft (such as aircraft carriers and submarines), engineering hardware, and some scientific research equipment (such as the supercollider and large lasers). A large residential or retail building may consume several megawatts in electric power and heating energy. The productive capacity of electrical generators operated by utility companies is often measured in MW. Modern high-powered diesel-electric railroad locomotives typically have a peak power output of 3 to 5 MW, whereas U.S. nuclear power plants have net summer capacities between about 500 and 1300 MW.[2]
What possible use do they have for a computer that chews up as much power as large retail building, a supercollider, or a diesel-electric freight locomotive? How much of this power goes up in wasted heat?
They simulate degradation of nuclear weapons (primarily), that's the use. Considering that the computer is 50x100 meters, it's pretty much the same size as a mall or large residential house. I don't know if it's the computer or computer+cooling that draws this ammount of energy, probably the latter, but the MW article paints a pretty nice piture of how freaking large and powerful this computer is, comparing it to aircraft carriers, supercolliders and lightning strikes. Almost every watt is wasted as heat, that's the way computers work, all they do is pushing electrons around, they don't do much in the mechanical field. Hard drives spin, but considering what goes into the processors and cooling, hard drives are neglible. As processors heat a machine, cooling must spend an comparable ammount of power to cool it. Ideally all electrical power should go to computing power, but in reality this is actually one of the hardest parts to do right when designing any computer today. Roadrunner has a performance/power ratio of 350 Megaflops/Watt which is really, really good, easily ranking as one of the most power efficient supercomputers (or any computer) out there. The reason for this is the tremendous power of the PowerXCell processors (compared to power requirements). Opterons are really crap in this respect. -- Henriok (talk) 13:13, 13 June 2008 (UTC)
BTW, a reference is needed for the 3.9 MW figure.

Timothy Perper (talk) 05:09, 13 June 2008 (UTC)

It's already referenced. Every figure in the "technical specification" section comes from the same source, reference [8] - RR Seminar - System Overview. -- Henriok (talk) 13:13, 13 June 2008 (UTC)
Did you mean reference [7] or reference [8]? I assumed that you meant reference 7:
7 ^ a b c d e f RR Seminar - System Overview. Los Alamos National Laboratory (2008-03-13). Retrieved on 2008-05-31.
8 ^ Los Alamos computer breaks petaflop barrier. IBM (2008-06-09). Retrieved on 2008-06-12.
You picked the right one, the number changed after I wrote the comment. I anticipated that, so I also supplyed the name of the referenced document. –– Henriok (talk)
So I added these two references to the article. The reason is that a reference for one paragraph does not carry over to the next paragraph unless it is explicitly annotated since it's quite possible that the technical data came from another source. If this is wrong, fix it and put the correct reference in.
Timothy Perper (talk) 16:21, 13 June 2008 (UTC)

On Topic[edit]

This article is about a individual supercomputer. It's not an article about general computing, it's not an article about why USA needs so simulate degradation of nuclear weapons, how and why it's not an article about defining what a supercomputer is nor an article about defining physical units, mathematical anotation, concepts in computer science or high performance computing in general. This is not an article that should describe parts of an computer and computer network in any detail, similarly, it should not describe why and how one should develop applications for a system like this or the general the ins and outs of supercomputing. This is not an article discussing the interconnects between components, nor discussing processor or memory architectures. And so forth.. Stay on topic people! There is plenty of room discussing all these other topics on their respective pages. If they are lacking, enhance those pages. It's not a fault of the article about Roadrunner if the page describing clustered file systems leaves much to be desired.

There's a lot stuff that might enhance this article, that _is_ on topic, such as describing how it is built, top-down, the other way of both. Illustrations might help too. There's a lot of illustrations in the reference materials, and they are there to be read, not just as proof that the data on this page is correct. If you want to know more, in detail, please consult the references, or liked pages on other topics.

Wikpedia is not a finished product, so there are probably tems and jaron that might need to be explained, but it's not a topic for this page, but another. This article is about a supercomputer. Stay on topic! -- Henriok (talk) 13:58, 13 June 2008 (UTC)

Once again, I don't agree. Roadrunner is not just another boring old supercomputer -- it's a showpiece for IBM, DOE, and computer science in general. It's also immensely expensive in construction and running costs (3.9 MW is not cheap). In fact, Roadrunner is sufficiently famous for it to have made it onto Wikipedia's home page! That fact alone means that we do not have the techie's luxury of imagining that we can talk only to other techies. You're talking to the general public. And judging from the number of people who have commented on this talk page that the article is incomprehensible, jargon-ridden gobbledegook (my word, not theirs), this article needs to be revisioned and rewritten.
Technical material needs to be explained to people like me, who are familiar with the general principles of computers and who are interested in this particular contraption. I noticed from your home page, Henriok, that you're Swedish, yes? You might not then know that in the US people -- not just computer techies -- are interested in these gadgets and want to understand them, and also want to know why our government is spending so much money to make one and only one computer. Similar (educated) public enthusiasm and interest surrounded the Hubble telescope, and a magazine like Scientific American sells very well to precisely that educated public. Since our clientele on Wikipedia includes many people in the US (although not exclusively!), anyone who writes for Wiki, especially on technical topics, has to be able to address questions from the general public. I suspect, though I do not know, that many European nations, including Sweden, have similarly interested publics -- and as long as Wiki is a general encyclopedia, and not a techies' bulletin board, we have to write our essays and entries to reach that public.
I am trained as a scientist -- I have a doctorate in biology (genetics) and have published in the appropriate scholarly literature in genetics and other areas (and I've been doing that for some decades; my PhD is from 1969). Those papers and books (yes, I have written several books) are directed to my fellow professionals, and not to the general public. But I also have other interests -- Japanese culture is one, and I've published about that too, as well as contributed to Wiki on the subject of Japanese pop culture in the US. In those articles, I try very hard to avoid jargon -- anyone with any knowledge of Japanese can snow an audience by throwing around incomprehensible Japanese words and phrases, but in my opinion Wikipedia does not need that kind of oneupsmanship, the kind where the author lords it over everyone else because he knows so much more (or thinks he does) than everyone else. Writing for an encyclopedia is the art of explaining -- it's not a matter of getting one up on everyone else. It has more in common with teaching than with impressing your colleagues with how much you know, or think you know.
So I'm asking you to get off your high horse and understand that this is Wikipedia, not a seminar on things that all advanced professionals in supercomputing (which I assume you are) already know and have known for years. That is not the appropriate level, IMO, for addressing the audience that Wikipedia has in the US, in Europe, and elsewhere in the world. If you answer that you do not want to write down for such an audience, well, that's fine -- then don't. But, by the same token, don't tell people who are trying to explain computers or any other aspect of science that we are doing it wrong. As you can tell from the other comments on this article, I am not the only person discontented by the level and focus of this article. In my opinion, Wiki should and must do better than this article has so far achieved.
You have a large role to play in that effort if you want it. You are an expert, and that means sharing your knowledge at a level that communicates with the general public. You might not want to do that, and that is your choice. You might want to take a far more elite view that scientists and technologically-trained people can and should ignore the great unwashed public, to quote a cliche, but I do not agree. I'm old enough and experienced enough to know that (a) we are ALL part of that public in areas we know little or nothing about and (b) that we have the responsibility -- yes, a strong word -- to make our knowledge available to other people. So I hope you can join and contribute to giving this article a better outreach to people everywhere in the world.
I hope that's clearer.
Timothy Perper (talk) 16:00, 13 June 2008 (UTC)
I still don't agree with you and I can't find _any_ article on Wikipedia that goes into the kind of detail that you seem to wish for. There are terms and jargon of every page, that's not described there, but linked to another page. Wikipedia is web of cross references for a reason. If one term is unclear, then provide a link to a page describing that page. This is happening on all pages on Wikipedia. So, if you want to continue tilting at windmills, please go ahead. I'm going to continue adding missing information to this page, not filling it with off topic fluff. Maybe you should too. –– Henriok (talk) 23:39, 14 June 2008 (UTC)
Please do. And I will continue to add criticisms. For an example of a wikipage that deals with a technical topic and is very clear, see the entry for "watt". It's one you yourself cited (as MW or megawatt), so I'm not looking very hard. I added back the {{Cleanup-jargon|date=June 2008}} that tag you removed. It's still jargon and it's still not clear. The two figures you added are (a) too small to see and (b) incomprehensible for the layperson. I'll leave them in and ask other people to say if they find them clear. You're still writing for an audience that doesn't exist on Wikipedia. And that, IMO, is a waste of time. BTW, citing other, equally incomprehensible articles from Wiki doesn't help your argument: it just proves that there are other articles on Wiki that need a lot of work. Timothy Perper (talk) 13:02, 15 June 2008 (UTC)
Are you serious? Should we define "watt" and the prefixes "Mega" in this article? Are you really, really serious? -- Henriok (talk) 16:37, 15 June 2008 (UTC)
This comment is inserted to follow Henriok's insertion. I think you're a bit confused. The article on watt -- which is what I'm talking about -- does define watt, and that's the right place to do it. So your claim that there aren't any Wikipedia articles that define technical terms for the layperson is false. More below. Timothy Perper (talk) 20:34, 15 June 2008 (UTC)
From TP. Here's another nice example. If I'm tilting at windmills asking for clarity, then looks like I'm in good company -- world-class company, in fact. Here's a quote from some LANL material on Roadrunner:
"A “flops” is an acronym meaning floating-point operations per second. One petaflop/s is 1,000 trillion operations per second. To put this into perspective, if each of the 6 billion people on earth had a hand calculator and worked together on a calculation 24 hours per day, 365 days a year, it would take 46 years to do what Roadrunner would do in one day."
That's clear, short, and direct. If they can do it, so can we. I can find more if you want. Timothy Perper (talk) 15:18, 15 June 2008 (UTC)
That's a comparison that's wrong! It doesn't cite what calculator and what calculation, and no calculator could ever do what Roadrunner does, no matter how long we tapped on one or how many people are tapping. So this figure is just pulled from a hat. It just paints a pretty picture that's false. Shoud we lie and make up stuff on Wikipedia just to dumb down the text so the general public thinks they understand the article? I'm just about calling names again. -- Henriok (talk) 16:19, 15 June 2008 (UTC)

Remain calm, Henriok. I'm not calling you names, and I don't think you're calling me names. Instead, a LANL source is using an example that is clear, if not perfect, to explain something. If you disagree with LANL, email them and complain. If the example is seriously wrong, then correct it. If it's false, prove that with the appropriately cited references. But I don't think you can claim that there are no examples on Wiki that can be comprehended by the general public. I have more comments below, under the item about the images. Timothy Perper (talk) 20:34, 15 June 2008 (UTC)

Where did the new images come from?[edit]

Henriok just added two new images of the system design for Roadrunner, which is useful -- the images are too small to see, but at least it's a step in the right direction. The copyright permission item for the first of these images says "I created this work entirely by myself" and is dated 2008-06-15, signed "Henriok (talk)". Well, OK -- so I pulled up the IBM technical reference (reference 7, except it's actually not an engineering description at all -- it was written for the general, educated public). It's dated March 13, 2008 and written by Ken Koch, identified as Roadrunner Technical Manager with the website as given in reference 7 of the Wiki article. So I downloaded it. Slides 21 and 22 sure look a great deal like the first of the two images Henriok says were done by him entirely by himself. No references were given in the captions of the two Wiki images from Henriok (they should be referenced to something). Entirely by yourself, Henriok, or did you, shall we say, borrow extensively from reference 7? If so, that needs to be credited, for example, by changing the caption by adding "Source: reference 7" with a modification in the Wikipedia copyright permission item for the image. At the very minimum, the two new figures need citations to sources. (The principle here is that if you draw something based on someone else's work, you have to cite the other person, and not claim that your drawing was done "entirely" by yourself.)

The material contained in reference 7 is **very** clear. It's technical, but it was written with precisely the kind of non-expert audience in mind that I've been saying the Wikipedia article has. So it's not tilting at windmills to ask that the Wiki article be rewritten at least as well as the IBM-LANL piece cited in reference 7. If they can do it, so can we. I won't and can't do that rewrite; I am not a technical expert. But someone else can, and I think should, rewrite this article.

Timothy Perper (talk) 13:59, 15 June 2008 (UTC)

OMG! Enough already! The stuff in reference 7 is hard even for me! I took a look at that presentation and I really found it confusing, so I took the figures already on Wikipedia (condenced but very accurate), and what I know about Opterons, Cell processors, LS21 and QS22 blades. I won't reference all sources that I have read to accumulate everything I know to make these pictures, such as reading IBM documentation and pressreleases for years and the various tech related information existing in this world. I swear that I was entirely alone at my computer when I did them! I did however consult documentation regarding LS21 and QS22. Please report my pictures for plagarism if you want to have then removed. Please edit the article to enlarge the pictures if you are unable to click on them to see them in a larger view. -- Henriok (talk) 16:32, 15 June 2008 (UTC)
To make things very, very clear. There's not a pixel, stroke, carachter or other property in those pictures that was not made with the assistanse of my computer and software by me, only me an no one else but me. They are original pictures made in their entirety by me based on knowledge of the ways of the world that I have aquired from a myriad of sources. If they resemble other pictures it's probably because both pictures depict something that was properly described to both artists. -- Henriok (talk) 16:52, 15 June 2008 (UTC)
They still look very similar to the images in the IBM/LANL document cited in reference 7. Plagarism is your word, not mine. I merely want to know where you got the images and the information they contain. If you insist that they are entirely, completely, and 100% your own work, then they're original research and they don't belong on Wiki. You can't have it both ways. You don't have to cite everything, just the parts necessary to convince your readers that you're not making it up. Timothy Perper (talk) 20:46, 15 June 2008 (UTC)
Then you might do some reading in the Wikipedia policy page about Original images. -- Henriok (talk) 22:28, 18 June 2008 (UTC)
This discussion isn't going anywhere. Henriok, you add anything you like to the article, with or without references. If it doesn't bother you, I'm not going to let it bother me either. It violates bunches of Wiki policies and makes nonsense of the basic principle of Wikipedia, that encyclopedia content has to be verifiable. But if that's what you want to do, I'm not going to stop you. Timothy Perper (talk) 21:31, 15 June 2008 (UTC)

From TP: I'm going to call it quits with this article. Henriok and I have irreconcilable differences that come from very different perspectives about what Wikipedia is and should do. Both of us are operating in good faith, but I have other things to do than follow this article or argue about it. I hope it works out, and I hope that someone comes along and clarifies the obscurities and opacities. But I'm taking this article off my watchlist. Timothy Perper (talk) 05:33, 16 June 2008 (UTC)

This computer may be powerful, but...[edit]

Can it run Crysis?-- (talk) 21:51, 21 July 2008 (UTC)

Yes it can. -- Henriok (talk)
Lol crysis on the roadrunner at full settings, DX10, resolution at 2056x1600 at 60fps? xD I gotta see that!!—Preceding unsigned comment added by (talk) 21:13, 14 September 2008 (UTC)
Hardly. The GPU is probably real sucky (ATI RN50) and you can only run it on one of the service nodes. It'll probably handle like a low end PC. Move along, these are not the gaming rigs you are looking for. -- Henriok (talk) 22:02, 14 September 2008 (UTC)
You can always use Nvidia Quadro lol, or GTX280 in tri-SLI :P —Preceding unsigned comment added by (talk) 00:25, 18 September 2008 (UTC)
This computer can not, and does not. -- Henriok (talk) 19:56, 19 September 2008 (UTC)

Stop saying nonsense here, go to a forum. -- (talk) 19:13, 1 November 2008 (UTC)

Design structure?[edit]

Info needed - why was this machine designed as it was? What is the logic behind its hybrid structure? FT2 (Talk | email) 22:06, 14 October 2009 (UTC)

I can't find any good source, so I can't write about it since it would fall under "original research" and "point of view" from my part. But.. I can answer your question here.
It's a best of both worlds kind of solution. Generic CPUs like Athlons are widely known, easy to program for, and are cheap, but not particularly powerful and draws a LOT of power. Not best at anything but good at everything. Specialized processors have superior performance but are often hard to program for since there's a certain level of obscurity involved. And they are expensive compared to commodity CPUs due to specialized development and low volume production. The Cell processors were designed to be a cross breed of a generic CPU and a specialized accelerator (best at certain things but only OK on all things). That would usually spell "expensive" but since Sony uses them in their PlayStation 3, they are cheap and they are not obscure. Development tools are readily available and researchers can buy a Roadrunner "prototype" for a couple of hundred dollars. Why not build a supercomputer entirely of Cells then? That is being done elsewhere, but this machine is so immense that there's a certain level of housekeeping necessary to keep performance optimal. Organizing workloads, monitoring, passing data around, collecting results, balancing workloads and such, and ofcourse do regular number crunching as well. Cell CPUs are not good at that, but Athlons are. Two different CPUs for two different tasks. The result is an unrivaled bang for the buck and best of breed power consumption. Many times more efficient than a traditional homogenous design. And,since there are a shitload of regular CPUs in Roadrunner, it's a formidable "regular" supercomputer even if one were to ignore the Cell accelerators.
Hybrid designs are common in this field. Cray's XT3/4/5 series of supercomputers work in a similar way, but they are using Athlons as the main number crunchers and custom made PowerPC processors called SeaStars to pass data around the thousands of CPUs. Cray are actually emphasizing the SeaStars more than the Athlons. This task is very important if you want to make this kind of computers balanced and efficient. RIKEN MDGRAPE-3 is another powerful hybrid supercomputer, with custom accelerators, actually delivering performance that surpasses Roadrunner on certain tasks. The coming years, supercomputers using GPUs as accelerators will be common, maybe even dominant, delivering energy efficiency, and bang for the buck that's an order of magnitude better than current homogenous designs. -- Henriok (talk) 19:27, 9 November 2009 (UTC)


As the intro says, its "Currently the world's second fastest computer", so whats the first? Seems like it might be an idea to include it--Jac16888Talk 14:12, 6 March 2010 (UTC)


What is the correct price: $133M as the article text says or $125M as the infobox says ? -- Juergen (talk) 22:50, 3 March 2013 (UTC)

April Fools Edit?[edit]

Anyone else feel this edit about "shredding" from a single source on April 1st seem, well, suspicious? For one, ArsTechnica is the only source and it is an article posted on April 1st. Second, "contributor" is new to this article and did a many edits. At least wait until April 2nd or a 2nd verifiable source for confirmation. --Mofoq (talk) 06:24, 2 April 2013 (UTC)

Possible source[edit]

  • "'Petaflop' supercomputer is decommissioned". BBC. April 1, 2013. Retrieved April 4, 2013.
Chris857 (talk) 02:55, 4 April 2013 (UTC)

External links modified[edit]

Hello fellow Wikipedians,

I have just modified 2 external links on IBM Roadrunner. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

As of February 2018, "External links modified" talk page sections are no longer generated or monitored by InternetArchiveBot. No special action is required regarding these talk page notices, other than regular verification using the archive tool instructions below. Editors have permission to delete the "External links modified" sections if they want, but see the RfC before doing mass systematic removals. This message is updated dynamically through the template {{sourcecheck}} (last update: 15 July 2018).

  • If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
  • If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers.—InternetArchiveBot (Report bug) 01:02, 10 November 2017 (UTC)