This discussion has been archived.
No new comments can be posted.
100 Teraflop Cray to Use Opterons
Related Links Top of the: day, week, month.
Two can Live as Cheaply as One for Half as Long. -- Howard Kandel
Two can Live as Cheaply as One for Half as Long. -- Howard Kandel
The mighty have fallen (Score:5, Informative)
Might want to check out what Cray and Sandia (Score:5, Informative)
Cray and Sandia say it is a 40 tera*OP* system, not a 100 teraflop one. See what Cray says here [cray.com] and what Sandia says here [sandia.gov] The really interesting thing is not the processor, but rather the interconnect which seems to be very similar to the torus used in the T3E [cray.com].
In other supercomputing news, check out what NERSC [nersc.gov] is proposing for their Earth Simulator Response Proposal [nersc.gov]. It's a 160 teraflop machine...
additional link (Score:2, Informative)
Re:UT (Score:4, Informative)
About 10.
UT (and every other game out there right now) assigns itself (its processer affinity) to the first CPU available. It makes absolutely no use of SMP or parallel execution whatsoever.
So it'll run on CPU 0, and the other 90 bazillion will sit idle.
Interestingly enough, all the gamer kiddies saving up for a shiney new Prescott based P4 with hyper-threading will see no advantage either, for the same reason.
AFAIK, Doom III will actually make use of concurrently executing threads, and there's rumor of a new UT2k3 exe that will, as well.
Programming for parallel CPU's is a whole new ballgame, and the rules are still being written.
Re:OT but... (Score:2, Informative)
Teraflop? (Score:4, Informative)
Clue: The s in teraflops is not a plural.
Re:Heating issues? (Score:5, Informative)
for the part of 'plasma cooling', it's similar (in non-scientific term) to laser cooling, which relates to absorbing momentum. (you may want to find some information on the plasma section of http://www.arxiv.org/ if you want to know 'bout that.)
WSJ story here (Score:5, Informative)
AMD's New Opteron Chips
Are Tapped for Red Storm
By DON CLARK
Staff Reporter of THE WALL STREET JOURNAL
Endorsing the technology of one of Intel Corp.'s key rivals, Sandia National Laboratories and Cray Inc. plan to build a massive supercomputer using a soon-to-be-introduced line of microprocessor chips from Advanced Micro Devices Inc.
The development project, estimated in June to cost $90 million, is a high-profile vote of confidence for AMD's new Opteron chip, in a small but prestigious market long dominated by other chip suppliers. It represents a missed opportunity for Intel, which has been targeting its new Itanium line at high-performance computing applications.
Red Storm, Sandia's name for the new supercomputer, also marks a step forward for the U.S. effort at leadership in supercomputers, which suffered a blow this year with the completion of a huge machine called the Earth Simulator by Japanese government agencies and NEC Corp. Where recent U.S. machines have largely been constructed out of components used in commercial computers, Cray is expected to develop special technology for connecting the AMD chips that should make Red Storm suited for more-complex scientific problems.
"This is a move away from commodity components," said Horst Simon, division director of the National Energy Research Scientific Computing Center, a supercomputer facility affiliated with Lawrence Berkeley National Laboratory. "It's very exciting."
Sandia, which does research for the U.S. Department of Energy in Albuquerque, N.M., and Livermore, Calif., has a performance goal of 100 trillion operations per second for Red Storm. It hasn't disclosed most technical details, including the chip selection. But Mr. Simon estimated that the machine will require 16,000 or more microprocessors to hit its speed target, which would appear to surpass the Earth Simulator's current performance.
Sandia said in June that it had selected Cray, a longtime supercomputer maker based in Seattle, to negotiate a development contract. Cray and Sandia officials didn't return calls seeking comment Friday. AMD and Intel officials declined to comment.
AMD could use some good news. The company's Athlon chip line, mainly used in personal computers, has been falling behind the performance of comparable Intel chips. The company reported last week a third-quarter loss of $254 million on sales of $508.2 million, off 34% from the year-earlier period.
Opteron is a high-end member of the new line, code-named Hammer, that is due out next year and viewed by analysts as AMD's best hope for recovery. Like the Itanium, Hammer chips are designed to process 64 bits of information at a time, instead of 32 bits, a capability that helps run huge databases and solve scientific problems.
Intel's Itanium line, developed over eight years with help from Hewlett-Packard Co., is based on an entirely new architecture and achieves its best performance on new 64-bit programs. AMD, by contrast, made 64-bit additions to the original Intel technology used in the past by both companies.
The difference, AMD says, allows Hammer-based computers to run both 32-bit and 64-bit software at high speed. AMD released preliminary test results last week for Opteron -- so far not validated by outside researchers -- that show the chip exceeding Intel's latest Itanium 2 model on one of two widely-used speed measures, AMD said.
Itanium 2, introduced last summer, has already been selected for at least a half-dozen high-performance installations. The Pacific Northwest National Laboratory, another Department of Energy facility, is building a $24.5 million system based on 1,400 Itanium 2 chips. Based on past Sandia announcements, the Red Storm project's stated performance goal is more than 10 times that of the Pacific Northwest project.
Write to Don Clark at don.clark@wsj.com
Re:Free-as-in-Beer (Score:5, Informative)
To differentiate, many postings here at
Re:Heating issues? (Score:5, Informative)
The Cray 1 and Cray X-MP were cooled by a freon-cooled cold plate. The Y-MP, C90, T3D, and T3E have a chilled liquid called Flourinert (some derivative of an artificial blood plasma, I believe, which is made by 3M) cirulating through a cold plate between boards. The Cray 2 and Cray T90 were cooled by being immersed in a vat of Flourinert. The Y-MP/EL, J90, and SV1 are all air cooled. The X1 (aka SV2) is cooled by spraying Flourinert onto the chips.
I believe, though I'm not 100% certain, that this system will be air cooled, presumably by lots of big fans
Re:Free-as-in-Beer (Score:2, Informative)
This is to differentiate it from Free-as-in-Speech.
The first usage of Free is to mean "Gratis" where as the second means "liberty." English just uses the same word for both concepts.
GNU stuff takes the "Liberty" meaning as being more important than the beer meaning. That is why you can copy a Redhat ISO without any problems. You are at Liberty to do so. Of course you can also pay Redhat if you want to.
Re:Heating issues? (Score:5, Informative)
Found this on the web -
http://www.cs.uu.nl/wais/html/na-dir/computer/syst em/cray/faq.html
"Keeping it cool - The development of Cray cooling technology allowed each technology generation to increase the circuit board density.
"Someone (perhaps Gary Smaby? I truly don't remember) once said that Cray Research was primarily a refrigerator company."
Cray-1: Single sided boards clamped to copper plates placed in aluminium racks that had cooling fluid in tubes.
XMP: Double side sandwich boards clamped to twin copper plates placed in aluminium racks which had cooling fluid in tubes.
Cray-2,3,4: Immersion cooling. The CPU and memory boards sat in a bath of electrically inert cooling fluid.
YMP, C90, T3d LC, T3e MC: Double-sided circuit boards clamped to hollow aluminium boards in which the cooling fluid circulated.
El,J90,T3eAC,SV-1: Blown air cooling.
T90: Immersion cooling. The CPU and memory boards sat in a bath of electrically inert cooling fluid."
I was up close and personal with an older Cray once - it was basically a tower of CPUs and very very short cables - and a whole bunch of cooling "units" srounding it. They were built into something like bench seats - the tech that was showing us around said they put those in so the guys could sit down and rest once in a while in peace :).
Duke
Re:Heating issues? (Score:4, Informative)
another is silicon-on-insulator (soi) where a layer of glass insulates each layer of silicon from the next. this allows lower voltages to be used because there is less interferance.
but even after all of that the opteron might run cooler than your athlon, but probably still run hotter than an intel chip at the same clock speed. check out the article on www.tomshardware.com (to lazy to look up the link) and take a look at the basic opteron heat sink requirements. must have copper contact area. bolts to mobo, not to socket (probably to cover for the added weight of the copper).
but that's just the regular opteron set-up.
wait, i forgot, amd fire comments are funny, right? maybe you didn't want discussion. too late now.
Re:Return of Vector Processing (Score:4, Informative)
Better reparse what he said. it uses the same design philosophy, not the same architecture. The X1 and Red Storm are distinctly different machines.
More of what you are worried about is this [computer.org]. That might be both scary and fun to code for.
However, it looks like vector processing is on the upswing, not down. It hit rock bottom during the 90s...
Re:Teraflop? (Score:5, Informative)
But usage has transformed the acronyms into words -- "teraflops", for example, has come to mean "more than one teraflop", where a "teraflop" is "one trillion floating point operations". The "per second" is now implied, for the main reason that the second is the metric unit of time, and the most common gauge for bulk computer operations.
De facto, usage has turned "TERAFLOPS" the acronym into "teraflop" the word (same for "megaflop" and "gigaflop").
Proper link (Score:4, Informative)
glad cray ditched the itanium (Score:4, Informative)
Remember that the customers who purchase these bad boys hire their own software engineers and purchase specialized compilers for maximum optimization. All the compilers will be available for Amd hammer chips because they run on so many systems. Also more engineers know it inside and out and can write great optimized programs for it.
Re:Heating issues? (Score:3, Informative)
The reason why this is a good thing is that modern processors do not generate their heat evenly, so you might have the bottom left 25% of the chip generating 50W, while the other 75% of the chip is mostly idle, only generating a few watts here or there. This is a bad thing for a variety of reasons, beyond the obvious fact that one area has to be cooled more than others. When it comes to actually dissipating the heat though, what you want is to get this heat into the furthest corners of the heatsink as quickly as possible. This is why copper inlays are a good thing for heatsinks (they spread the heat away from the hot core towards the edges of the heatsink faster than aluminium does), however even aluminium conducts heat a lot better than the P4's heat spreaders do.
This sort of thing is, presumably, less of a problem for the Athlon as compared to the P4 simply due to die sizes. The P4 is a MUCH larger die (nearly twice the size) when compared to the Athlon. So, while the distance between the hottest and coldest points in an Athlon are always going to be quite small, they can be relatively large on a P4.
As for fixing the heatsink to the motherboard instead of the socket, this is a damn good thing if you ask me. AMD originally thought that this would be how things would be done with the Athlon, and specified 4 holes around the socket to do just such a thing. You can even buy heatsinks that attach this way (my Athlon is cooled by just such a heatsink, an Alpha PAL 8045), however the vast majority of heatsink designers just flat out ignored these holes and just attached to the heatsink. Intel was a bit more forcefull, and absultely required ALL heatsinks to attached either to the motherboard or even to bolt to the case itself (the first P4s, those that came in a socket 423 package, required a special case that the heatsink bolted on to). Intel's currently design of using clips to clamp the heatsink down to the motherboard is a fairly good one IMO, though it does put an awful lot of strain on the board. AMD's plan for the Hammer/Opteron is similar, and it looks like they're going to try to avoid some of the motherboard bending/strain, however it remains to be seen just how well it will work in practice.
In any case, it remains to be seen just what the power consumption of the Hammer processors will be. The only thing that is for certain is that they will use a LOT less power than their Intel counterparts (which, in this particular case, are the Itaniums, which use roughly twice as much power as the hottest running Athlons). IBM's Power4 chips also pump out huge amounts of heat, so relative to these competitors chips, AMD's Opteron could be a fairly low-power solution.