[Beowulf] Multicore Is Bad News For Supercomputers
coutinho at dcc.ufmg.br
Fri Dec 5 09:09:12 PST 2008
2008/12/5 Robert G. Brown <rgb at phy.duke.edu>
> On Fri, 5 Dec 2008, Eugen Leitl wrote:
>> (Well, duh).
> Good article, though, thanks.
> Of course the same could have been written (and probably was) back when
> dual processors came out sharing a single memory bus, and for every
> generation since. The memory lag has been around forever -- multicores
> simply widen the gap out of step with Moore's Law (again).
> Intel and/or AMD people on list -- any words you want to say about a
> "road map" or other plan to deal with this? In the context of ordinary
> PCs the marginal benefit of additional cores after (say) four seems
> minimal as most desktop users don't need all that much parallelism --
> enough to manage multimedia decoding in parallel with the OS base
> function in parallel with "user activity". Higher numbers of cores seem
> to be primarily of interest to H[A,PC] users -- stacks of VMs or server
> daemons, large scale parallel numerical computation.
Datamining is useful for both commercial and scientific world and is very
data-intensive, so I think this issue will be adressed, or at least someone
(Sun, for example) will build processors for data intensive applications
that are more balanced, but several times more expensive.
> In both of these
> general arenas increasing cores/processor/memory channel beyond a
> critical limit that I think we're already at simply ensures that a
> significant number of your cores will be idling as they wait for
> memory access at any given time...
>> Multicore Is Bad News For Supercomputers
>> By Samuel K. Moore
>> Image: Sandia
>> Trouble Ahead: More cores per chip will slow some programs [red] unless
>> there's a big boost in memory bandwidth [yellow
>> With no other way to improve the performance of processors further, chip
>> makers have staked their future on putting more and more processor cores
>> the same chip. Engineers at Sandia National Laboratories, in New Mexico,
>> simulated future high-performance computers containing the 8-core,
>> and 32-core microprocessors that chip makers say are the future of the
>> industry. The results are distressing. Because of limited memory bandwidth
>> and memory-management schemes that are poorly suited to supercomputers,
>> performance of these machines would level off or even decline with more
>> cores. The performance is especially bad for informatics
>> applications—data-intensive programs that are increasingly crucial to the
>> labs' national security function.
>> High-performance computing has historically focused on solving
>> equations describing physical systems, such as Earth's atmosphere or a
>> hydrogen bomb's fission trigger. These systems lend themselves to being
>> divided up into grids, so the physical system can, to a degree, be mapped
>> the physical location of processors or processor cores, thus minimizing
>> delays in moving data.
>> But an increasing number of important science and engineering problems—not
>> mention national security problems—are of a different sort. These fall
>> the general category of informatics and include calculating what happens
>> to a
>> transportation network during a natural disaster and searching for
>> that predict terrorist attacks or nuclear proliferation failures. These
>> operations often require sifting through enormous databases of
>> For informatics, more cores doesn't mean better performance [see red line
>> "Trouble Ahead"], according to Sandia's simulation. "After about 8 cores,
>> there's no improvement," says James Peery, director of computation,
>> computers, information, and mathematics at Sandia. "At 16 cores, it looks
>> like 2." Over the past year, the Sandia team has discussed the results
>> with chip makers, supercomputer designers, and users of high-performance
>> computers. Unless computer architects find a solution, Peery and others
>> expect that supercomputer programmers will either turn off the extra cores
>> use them for something ancillary to the main problem.
>> At the heart of the trouble is the so-called memory wall—the growing
>> disparity between how fast a CPU can operate on data and how fast it can
>> the data it needs. Although the number of cores per processor is
>> the number of connections from the chip to the rest of the computer is
>> So keeping all the cores fed with data is a problem. In informatics
>> applications, the problem is worse, explains Richard C. Murphy, a senior
>> member of the technical staff at Sandia, because there is no physical
>> relationship between what a processor may be working on and where the next
>> set of data it needs may reside. Instead of being in the cache of the core
>> next door, the data may be on a DRAM chip in a rack 20 meters away and
>> to leave the chip, pass through one or more routers and optical fibers,
>> find its way onto the processor.
>> In an effort to get things back on track, this year the U.S. Department of
>> Energy formed the Institute for Advanced Architectures and Algorithms.
>> Located at Sandia and at Oak Ridge National Laboratory, in Tennessee, the
>> institute's work will be to figure out what high-performance computer
>> architectures will be needed five to 10 years from now and help steer the
>> industry in that direction.
>> "The key to solving this bottleneck is tighter, and maybe smarter,
>> integration of memory and processors," says Peery. For its part, Sandia is
>> exploring the impact of stacking memory chips atop processors to improve
>> memory bandwidth.
>> The results, in simulation at least, are promising [see yellow line in
>> "Trouble Ahead
>> Beowulf mailing list, Beowulf at beowulf.org
>> To change your subscription (digest mode or unsubscribe) visit
> Robert G. Brown http://www.phy.duke.edu/~rgb/<http://www.phy.duke.edu/%7Ergb/>
> Duke University Dept. of Physics, Box 90305
> Durham, N.C. 27708-0305
> Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu<email%3Argb at phy.duke.edu>
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Beowulf