Slashdot Mirror

← Back to Stories (view on slashdot.org)

200-Gigaflop Mac Cluster

Posted by pudge on from the i'll-buy-that-for-a-kroner dept.

Mauro Notarianni writes in that the Danish Technical University (DTU) has developed the Scandinavia's largest dedicated Mac cluster. "Velocity-X" is packed with 32 dual G4/800's (200 Gigaflops), and will be turned on Monday. Its primary use will be studying the influence of proteins on cancer, and, more importantly, large film and animation projects. It can be rented for DKK 50,000 (about USD 6,500) per week.

49 comments

  1. More importantly? by PeekabooCaribou · · Score: 1

    Its primary use will be studying the influence of proteins on cancer, and, more importantly, large film and animation projects.

    It's nice that the /. community is sensitive to a variety of issues.

    --
    "I'll say it again for the logic-impaired." -- Larry Wall.
    1. Re:More importantly? by Anonymous Coward · · Score: 1, Informative

      Start here!
      Continue here!

      By the way, why use regular dual 800, when there is Xserve?

    2. Re:More importantly? by RevGregory · · Score: 1

      Well, considering that this project was almost certainly begun BEFORE the Xserve announcement and that the Xserve isn't shipping as of yet - it would still be vaporware instead of a reality. Just my opinion!

    3. Re:More importantly? by Anonymous Coward · · Score: 0
      Outside of being graphic toys when have they ever really delivered in the business or scientific fields?

      I know you don't understand big words and all, but G4s were used to crack the human genome...

      The F-14 was designed on PowerMacs...

      NASA uses them for lots of things...

      I AM THE ANTI TROLL! BWAHAHAHA

    4. Re:More importantly? by Anonymous Coward · · Score: 0

      I know you don't understand big words and all, but G4s were used to crack the human genome...

      Also quite interesting, since Genentech Blast was announced this year, and the human genome was published january 2001.

  2. Importance by recursiv · · Score: 4, Insightful

    Its primary use will be studying the influence of proteins on cancer, and, more importantly, large film and animation projects.

    More importantly? Animation? Than cancer?
    Apparently someone hasn't had any relatives die of cancer recently. That's a pretty insensitive remark. If this was a comment, I would be sure it was a troll.

    --
    I used to bulls-eye womp-rats in my pants
    1. Re:Importance by Anonymous Coward · · Score: 0

      pardon my sarcasm in advance....
      thats kids today. Anyone who values high end film rendering over health science is simply ignorant and or too young and stupid to have an opinion worth reading. Once you lose someone to cancer, your priorites change. That geroge lucas can save a few bucks on episode 3, or some gigaflops used towards understanding protein chains...... which is going to matter more for humanity? Hmmmmm... guess its not sexy enough.

    2. Re:Importance by thetzar · · Score: 1

      -1, doesn't understand sarcasm.

    3. Re:Importance by Anonymous Coward · · Score: 0

      Thanks, I was going to comment on that. My own mother was diagnosed with thyroid cancer about 18 years ago or so and, thankfully, was completly cured. Of course she has no thyroid, though.
      Anyone who could say entertainment is more important that cancer research really needs to reoder their priorities.

    4. Re:Importance by Anonymous Coward · · Score: 0

      he's being sarcastic (monty python style). get the hint would you, ffs.

    5. Re:Importance by Anonymous Coward · · Score: 0

      wrong, you dont understand sarcasm. last line is sarcastic dipshit. go back to 7th grade english

  3. nice, but... by Anonymous Coward · · Score: 0

    ...there are many of these clusters around the U.S. This one was built in 1999 at Iowa State University/DOE. It consists of 48 G4 processors, runs Black Lab Linux, and was never given any attention by the press. This new cluster is just a more commercialized version of the same thing.

  4. Obligitory comment by xinu · · Score: 1, Troll
    Should I say it? I hate to. But I should. Cuz if I don't, someone else will. So here goes:

    "Imagine a beowulf cluster of theeeeese "

  5. Tell us on Monday by Anonymous Coward · · Score: 0

    will be turned on Monday.

    Until then it is a huge pile of plastic, silicon and cables.

    Tell us when it is working.

  6. I gotta say it:) by dcstimm · · Score: 0

    That is one GAY cluster! Its so fruity looking hence "APPLE". Very cool though I wish I had it in my basement. but they are probably pissed off that they didnt use the xserves they would have saved alot of space, and could have been stacked. Oh well....

    --
    keanmarine.com
    1. Re:I gotta say it:) by Anonymous Coward · · Score: 0, Flamebait

      hahahhahahahhahahahhahahahhahahahahhah ;-) ;-)

    2. Re:I gotta say it:) by Anonymous Coward · · Score: 1, Insightful
      That is one GAY cluster! Its so fruity looking hence "APPLE".

      So you are always thinking homosexual thoughts, eh? I bet you wish you could suck every cock on /. don't you? Grow up already faggot boy.

    3. Re:I gotta say it:) by dcstimm · · Score: 0, Redundant

      heh naw, but I can see that you do;-)

      --
      keanmarine.com
  7. Xserve? by owenc · · Score: 1

    I would imagine using the Xserve would take a lot less space

  8. questions by benjamindees · · Score: 1

    Is this cost-competitive with generic x86 clusters?

    Does this offer any advantages over generic x86 clusters?(rendering software, I assume?)

    What is a technical university going to do with this that requires Apple graphics software?

    --
    "I assumed blithely that there were no elves out there in the darkness"
    1. Re:questions by benh57 · · Score: 2, Informative

      The OS X version os BLAST (a bioinformatics package - what they are doing) is apparently many times faster than any PC version, due to its altivec/vector optimization.

    2. Re:questions by Anonymous Coward · · Score: 0

      you can tell some of these mac bigots that the g4 does some things better 1000 times and they just dont hear it.

    3. Re:questions by Anonymous Coward · · Score: 0

      1. They aren't doing bioinformatics, but quantum chemistry. This 'curing cancer' is actually bordering to a LIE just to get publicity. A quantum-chemistry calculation of a protein is limited to picosecond scales, and even the microsecond scales we can access with classical molecular dynamics aren't enough to understand proteins. You can calculate exact energies for things like oxygen binding in hemoglobin with quantum chemistry, but you will NOT understand cancer.

      2. The Genentech BLAST is 'up to' 5 times faster (no details given) than the free, unoptimized version from NCBI compiled with
      gcc instead of the faster intel compiler. It is still slower than commercial versions of BLAST for x86.

  9. MacOS X cluster by leastsquares · · Score: 2

    MacOS X doesn't strike me as a good choice for a cluster OS. No process migration. No kernel-level checkpointing. No network channel bonding. etc. etc.

    Does anyone know what software they'll be using to study those proteins?

    1. Re:MacOS X cluster by foo12 · · Score: 2, Informative

      This page is mostly Apple PR but it does link to several clustering solutions for Mac OS X as well as Mac OS 9.

      As for the software running over the cluster itself, I'd imagine most of it --- if not all of it --- is custom written for the task at hand.

  10. I found... by benjamindees · · Score: 2, Informative

    this article, which may shed some light on what they plan to do with this and why OSX is a requirement. Does anybody know anything more about protein crystallography and why it would require "stereo" video cards?

    --
    "I assumed blithely that there were no elves out there in the darkness"
    1. Re:I found... by Jeremy+Erwin · · Score: 4, Informative

      If you shine light through a crystal, it is diffracted in a specific manner that can be corollated with the internal structure of that crystal. Shining X-Rays through crystallized protein will similarly produce a diffraction pattern that can be used to determine the three dimensional structure of that protein.

      A stereo video card is a system that can present different video images to the left and right eye of the user, producing an illusion of depth. I would imagine that this would be very helpful in visualizing three dimensional protein structure.

  11. xserve not shipping yet by johnpaul191 · · Score: 2, Informative

    i do agree it seems like a much cleaner happier way to do it (if you have the money). unfortunately there are some issues.......

    1) xserve isn't shipping yet. apple site says 2-3 weeks still, though there are rumors they will be shipping any day now.

    2) the dual processor xserve is $4,000 (the single processor is the $3000 model). i would wonder if the included software would be seen as a waste when buying a plethora of these. then again if you are curing cancer i guess worrying about the extra $$$$ for the right equipment seems like crying over spilled milk.

    as posted above, dual 800mghz G4 machines are older stock so i am guessing they were either purchased a while ago, or they got a good price break. i think we'll see them down the road, supposedly the xserve will be easy to cluster. actually according to stories posted here and on MacSlash in the past it seems that people can unbox, setup and startup a G4 cluster in a day.

    1. Re:xserve not shipping yet by Hadlock · · Score: 1

      "1) xserve isn't shipping yet. apple site says 2-3 weeks still, though there are rumors they will be shipping any day now."

      likely the xserve doesn't run anything less than 10.2, or some variant of OS X server 10.2, which means they'll probably be arriving @ scientist's doorsteps the day of MWNY keynote.

      --
      moox. for a new generation.
  12. Impressive... by SvnLyrBrto · · Score: 2, Funny

    > The F-14 was designed on PowerMacs...

    Very impressive indeed.

    Even MORE impressive, considering that the F-14 was designed years before Apple even existed; and more than a DECADE before the Macintosh was created.

    I wonder how Steve pulled THAT one off?

    cya,
    john

    --
    Imagine all the people...
  13. Re:32 dual g4's! by Anonymous Coward · · Score: 0

    your full of shit, for what its doing BLAST, a pc cant beat a G4. do some home work and prove me wrong

  14. Sounds Great by Higatsuku · · Score: 3, Insightful

    Consistantly, Apple has been making strides to create a platform that is ideal for scientific applications like this. The Altivec processing unit allows the G4 to handle 256bit chunks of data at a time and process it fast. The large cache will allow the process to never starve for info from ram. Optimized number crunching aps for Mac smoke (most) anything else out there, and certainly anything on the x86 side of computing.

    Further more, anyone out there who complains about wintel and isn't up to their neck in open source development should back off on the Apple bashing. Here is a company who is, at the very least, trying to innovate or just evolve computing to be better for everyone. (while making money, as any business is designed to) Unintelligent Apple bashing is getting pretty old by now. Why don't you come up with some factual arguements..or better yet, go to an Apple store and use one for a while. Start the terminal..find out just how half-baked windows is compared to OS X and how easy the Mac OS is compared to the hair pulling fun of linux. (yes I know red had can be installed by a monkey, but if you break it, good luck for any newbie)

    Just a piece of my mind.

    1. Re:Sounds Great by Anonymous Coward · · Score: 0

      Optimized number crunching aps for Mac smoke (most) anything else out there, and certainly anything on the x86 side of computing.

      Sorry, but this just isn't true. As far as I know, there is one single application (Genentech BLAST) where a heavily altivec-optimized G4 version beats a completely unoptimized x86 version. What Apple don't say is that there are several commercial x86-optimized BLAST versions that beat the Genentech one.

      The true story is that anything using double precision (and most scientific/numbercrunching code does!) will run a factor two faster on x86 than G4, simply because Altivec only does single precision. There are also several applications optimized for double precision SSE2, while there is no such alternative on the G4.

      It is also a myth that Altivec should be so much better than SSE. Sure, it does better per clock, but the clock in a P4 is more than twice that of a G4. Just check out e.g. ATLAS or GROMACS. Both these programs use Altivec on G4 and SSE/SSE2 on x86, and the x86 performance is significantly better even in single precision, not to mention the hardware price.

      Finally, this assumes you don't consider the cost of porting the code to Altivec, or that writing it in Altivec from the beginning makes it unportable. This might not be important to photoshop users, but it is for scientific use: there isn't a single Altivec-capable machine among the world's 500 fastest supercomputers. If you just take a portable source and compile it, code generated by the Intel compilers on x86 will run about twice as fast as anything generated on G4.

      No, I'm not a 'mac basher'. But you don't serve the Mac community by inflating benchmark results (Apple "flops" vs. real double precision flops, comparing hand-optimized code vs. nonoptimized code, not giving any details on how benchmarks where performed, not publishing SPECbench results like ever other vendor does).

      All this will do is make people p*ssed off when they compile their first code on a G4 and realize it's a factor 2 slower.

      For a comparision, have a look at a recent AMD cluster installation at

      HPC2N. They could have used SSE/3DNow instructions to increase a bogus "flopcount" by a factor 4, but they don't. Instead they only mention the standard double precision flops (800 Gflop peak), and in practice they really achieve 481 Gflops on LINPACK.

    2. Re:Sounds Great by Higatsuku · · Score: 1

      Maybe they will just be saving money on power consumption? :) I can be a sucker for the apple propaganda in hopes that it can sweep intel/ms based stuff away.

      Why, do you suppose Genentech decided to go the PowerPC route if x86 can be so much faster?

      I've also heard that the x86 based processors don't work as well with multiple CPUs or in clusters as do the PowerPCs.

      Power4 machines are PowerPC/risc (although I could be wrong) ..that has to count for something. I don't think motolorla is putting down the money or time for clusters the way that IBM is.

      Additionaly, my comments weren't directed at objective/factual commintary, but at those who pass off apple as a passing fad or some joke of a computer. I personally use both a WinXP box and a Mac OS X laptop. (but the wintel box was free)

    3. Re:Sounds Great by Anonymous Coward · · Score: 0

      Power4 machines are PowerPC/risc (although I could be wrong) ..that has to count for something. I don't think motolorla is putting down the money or time for clusters the way that IBM is.

      Well, yes and no; they are actually called PowerPC now, but it's really IBMs PowerParallel architecture. In my opinion, IBMs choose a much better solution than Motorola; the PowerParallel chips have TWO extremely strong general-purpose floating point units, and beat everything else in the world when it comes to performance. You don't have to handcode any hardware-specific code at all - the compiler does everything for you.

      The Motorola PowerPC chips, in contrast, have an extremely weak floating-point unit and yields absolutely horrible performance unless you handcode for Altivec.

      Do get me right - I HAVE handcoded both Altivec and SSE in contrast to most people here, but neither is a good long-term solution since it breaks portability and incurs extra development time/cost. Five years ago everyone was using Alpha chips - we have no idea what we are going to use in another five years, but the probability that it will be identical to the current CPUs is small. This is one of the reasons why SPECbench is so popular among scientists; the vendor is allowed to do ANYTHING, but it must be generated automatically by the compiler.

  15. Re:32 dual g4's! by Anonymous Coward · · Score: 0
    your full of shit
    My full of shit what?
  16. Re:32 dual g4's! by Anonymous Coward · · Score: 0

    Yoda say, gay sex man are you?

  17. Re:32 dual g4's! by Lars+T. · · Score: 0, Offtopic
    My full of shit what?

    Skull. HTH. HAND.

    --

    Lars T.

    To the guy who modded me down from perfect to terrible Karma - Apple haters still suck

  18. that would be 200 GFlops of theoretical peak by dario_moreno · · Score: 2

    counting about 4 fp per clock cycle! Only a trivial loop in assembler can achieve this kind of performance. I bet that on the LINPACK benchmark, considering a fast ethernet network, this thing can achieve maybe 50 GFlops max depending on the amount of memory available. A cluster of XP 1800s delivering the same performance costs about $30,000 nowadays including a switch. Besides, I see only the point of paying for computer power in the case of, say, a 1 hour run on a 1000 processor machine with 500 Gb of RAM, which one could never achieve on "normal" workstations. The danish MacWulf doesn't seem so interesting.

    --
    Google passes Turing test : see my journal
    1. Re:that would be 200 GFlops of theoretical peak by @madeus · · Score: 2

      The danish MacWulf doesn't seem so interesting.

      Your right, I can't see why people reading apple.slashdot.org would be interested in reading about G4 clusters.

      :P

    2. Re:that would be 200 GFlops of theoretical peak by Anonymous Coward · · Score: 0

      Not even close...

      Sorry, but the industry standard way of ranking supercomputers is DOUBLE PRECISION floating point operations, not vectorized single precision ones. LINPACK is double precision, so the *theoretical peak* of this cluster is 25 Gflops. With 100 Mbps networking and non-DDR memory I would eat my left shoe if it in practice can achieve more than 10-15 LINPACK Gflops.

      Altivec is a wonderful bonus, but nobody is basing intel clusters on SSE performance, or Alpha, MIPS clusters based on their multimedia instructions.

      I'm working in the same field as these guys, and while we have plenty of Macs on our desks we wouldn't dream of basing a cluster on G4. Quantum chemistry operations are heavily dependent on the
      BLAS libraries. ATLAS is optimized for Altivec on G4 and SSE/SSE2 in intel, but even in single precision the pentium4 performs significantly better than the G4 at a much lower price. In double precision the G4 isn't even in the game. And very few people think single precision is accurate enough for quantum chemistry...

  19. Re:32 dual g4's! by Anonymous Coward · · Score: 0

    your full of shit. that is all. pretty self explanatory. If i wanted to go into detail and explain all the reasons your wrong, i would have. But becuase YOUR FULL OF SHIT, it would be a waste of both of our time. This is a much more economical way to debate a troll. Its always easier to go down to your level and your reply demonstrates you felt comfy with it too. have a nice day!

  20. Re:32 dual g4's! by usr122122121 · · Score: 0, Offtopic
    your full of shit. that is all.
    your --> you're
    --

    -braxton
  21. All I can say is.... by qurob · · Score: 1


    PowerPC's smoke in SETI....so I'm sure this setup is pretty cheap.