Forgot your password?
typodupeerror
Microsoft Supercomputing

Cray Co-Founder Joins Microsoft 169

Posted by CowboyNeal
from the switching-horses dept.
ergo98 writes "Burton Smith, co-founder and chief scientist at Cray (The Supercomputer Company), has jumped ship. He's joining Microsoft to help them with their clustered computer initiative. Burton joins Microsoft as a technical fellow."
This discussion has been archived. No new comments can be posted.

Cray Co-Founder Joins Microsoft

Comments Filter:
  • by RodgerTheGreat (905510) on Saturday November 26, 2005 @11:55AM (#14118831)
    and computing power. Before I get on a rant about the megahertz myth and why I love PowerPC's, the real reason Crays were powerful was their massive parallelism and the use of path optimization (premeasured cables and careful curcuit designs that made the distance electrons had to travel equal between parts of the machine) was the real reason they were a Cray.

    Just because your machine is *faster* doesn't mean it's anywhere near as powerful! How many CPU cores does your machine have? I bet the cray had more. Clockspeed means *nothing*. The reason those applications don't exist is because they would take an order of magnitude as long to calculate on your "old computer".

    I recommend you do some reading on supercomputing-http://en.wikipedia.org/wiki/Superc omputing [wikipedia.org]

    "Supercomputers traditionally gained their speed over conventional computers through the use of innovative designs that allow them to perform many tasks in parallel, as well as complex detail engineering. They tend to be specialized for certain types of computation, usually numerical calculations, and perform poorly at more general computing tasks. Their memory hierarchy is very carefully designed to ensure the processor is kept fed with data and instructions at all times--in fact, much of the performance difference between slower computers and supercomputers is due to the memory hierarchy design and componentry. Their I/O systems tend to be designed to support high bandwidth, with latency less of an issue, because supercomputers are not used for transaction processing."
  • by Dread_ed (260158) on Saturday November 26, 2005 @11:58AM (#14118840) Homepage
    ...wouldn't you just love to spend Bill's seemingly ulimited resources to fund your pet project?

    The guy is in the business of developing the biggest/fastest/floppiest computers he can. Having the deep-as-the-Pacific pockets of Microsoft to dig into can't hurt his chances of implementing all his pie-in-the-sky ideas.

    Smart move if you ask me.
  • by djupedal (584558) on Saturday November 26, 2005 @12:01PM (#14118856)
    Smith: "One more thing" [taborcommunications.com] is that the uniprocessor has pretty well run out of steam. Parallelism to date has been a nice strategy for HPC users and an afterthought for microprocessor vendors. Now, it is becoming a matter of business survival for all processor vendors. Parallelism is going to be taken more seriously, starting with the idea of exploiting multi-threading and multiple cores on a single problem. This is a major change. Imagine if Microsoft wanted to write Office in a parallel language. What would that language be, and what would be the architecture to support it? We don't have good answers to these questions yet'

    Imagine if you got paid to answer that question? Which, by the way comes out as 'parallel' and 'parallel language' (don't mix them up) ...the other shoe drops.
  • by deadline (14171) on Saturday November 26, 2005 @12:28PM (#14118966) Homepage
    This is kind of odd. Burton Smith is not really a cluster guy, although he probably knows his way around HPC (High Performance Computing). Cray is not really a cluster company [newsforge.com] (except for the system they bought from Octiga Bay [technewsworld.com] deal). If you want to read a review of what Bill Gates said at the recent Supercomputing conference, check out Where is the Cluster? [clustermonkey.net] at Cluster Monkey.
  • by sluke (26350) on Saturday November 26, 2005 @01:48PM (#14119288)
    I agree very much with your post, but would just like to point out that all of this really depends on your application. As an example, many high performance applications use the metropolis method to do Monte Carlo, and in that case (as in many other "embarrasingly parallel" applications) the interconnect hardly matters at all.
  • by itsNothing (761293) on Saturday November 26, 2005 @02:00PM (#14119351)
    The real reason that early Crays were powerful was because they were very fast (high speed devices), and their main memory was SRAMs (very low latency, but smaller in size) instead of DRAMs (high latency, large size), so memory requests were serviced quickly. A friend once said that a Cray was a great lisp machine because it had a low latency memory.

    The vector registers were interesting, but only of utility for linear algebra problems (Matrix operations), and then only when the vector sizes were fairly large. Their architectural contribution is overrated.

    The parallelism you describe is a result of the attack of the killer micros . There was no way that an innovative architecture could compete against the relentless advances in device technology pursued by Intel and others. Most of the modern tera-flop systems use oodles and oodles of "stock" micros and a high performance interconnect network.

    Today, the era of the killer micros is about over. Micro manufacturers can't just speed up their devices because they are already operating close to a limit of device technology. "Multi-core" processors are being built primarily because the manufacturers don't know what else to do. We're also about to start seeing the return of "architecture" to the realm of computing.

  • Re:Irresistable (Score:3, Insightful)

    by einhverfr (238914) <chris.traversNO@SPAMgmail.com> on Saturday November 26, 2005 @04:26PM (#14119968) Homepage Journal
    With all their cash, they can catch up in a big hurry. Also, with their market position they can bide their time. How long did it take NT/2000/XP to become somewhat respectable?

    Well....... People seriously misunderstand what it would take for Microsoft to take on Cray.

    I think it is very informative to read Cray's K-10 SEC forms. The describe in pretty good detail different types of supercomputers. Microsoft (because they are not a vertically integrated computer/software manufacturer) can only really do the MMP (Massively multiparallel) supercomputers well. Indeed, most of the software for MMP supercomputing projects, such as MPI and PVM is available open source (user mode). MMP work well for certain types of tasks, but they break down for others. For other types of tasks, you are better off with a supercomputer based on high-speed interconnects between processing units. These are not like beowulf clusters. For these types of supercomuters, Cray and NEC are pretty much the only game in town. For those with a short memory, NEC is a "baby bell" (an AT&T spin-off).

    So if Microsoft can't compete in the high-speed interconnect supercomputing market, what about the MMP? Can their products be competitive?

    My answer is "probably not." They have several serious issues to overcome. The first is that Windows licenses are far more expensive than Linux licenses so the cost per node is likely to be higher. Most MMP supercomputers today run either some varient of Linux or UNIX. Linux tends to be run on commodity hardware while UNIX is either run on some control nodes (in some of the Cray MMP offerings), or on big iron where there may still be some performance benefit.

    This therefore represents an attempt to take Windows and take on one of Linux's market strongholds. The best they can hope for is to be able to get market access to some of the peripheral units. Even there, I am not sure how successful Microsoft can be. Much of the work there is likely to depend on SFU because most of the supercomputer programmers are going to be more familiar with a POSIX development environment than an Win32 one. There might be a few VMS geezers out there who might find this helpful but in general, I just don't see it.

Our informal mission is to improve the love life of operators worldwide. -- Peter Behrendt, president of Exabyte

Working...