On Tue, Nov 27, 2012 at 06:17:55PM -0500, Prentice Bisbal wrote: > > On 11/27/2012 03:37 PM, Douglas Eadline wrote: > > > >> My interest in Arm has been the flip side of balancing flops to network > >> bandwidth. A standard dual socket (AMD or Intel) can trivially saturate > >> GigE. One option for improving the flops/network balance is to add > >> network bandwidth with Infiniband. Another is a slower, cheaper, cooler > >> CPU and GigE. > >> > > applause. > > I applaud that applause. > > What Bill has just described is known as an "Amdahl-balanced system", > and is the design philosophy between the IBM Blue Genes and also > SiCortex. In my opinion, this is the future of HPC. Use lower power,
The laws of the universe agree with your opinion. A provably optimal classical computing configuration requires a close packing of the nodes so that they're within each other light cones to minimize communication latency. A sea of nodes on a mesh is will approach that asymptotically, by shrinking the grain size until the units of memory and computation is identical. Notice that you don't need a global clock, as local communication can be asynchronous, and in general a locally weakly coupled system of oscillators will synchronize soon after bootup, until they beat in unison. Your problem is mapped to a virtual circuit trace, laid out in a 3d volume, with a volume refresh rate commensurable with geometry of the light cone, which has a physical limit of about ~nm. I also hope that spintronics, both for storage (MRAM) and computation will reduce the power envelope, and also increase reliability (e.g. spintronics is effectively immune to radiation). > slower processors, and then try to improve network performance to reduce > the cost of scaling out. Essentially, you want the processors to be > *just* fast enough to keep ahead of the networking and memory, but no > faster to optimize energy savings. > > The Blue Genes do this incredibly well, so did SiCortex, and Seamicro > appears to be doing this really well, too, based on all the press > they've been getting. With the DARPA Exascale report saying we can't get > to Exascale with current power consumption profiles, you can bet this > will be a hot area of research over the next few years. > > Okay. I'm done listening to myself type. _______________________________________________ Beowulf mailing list, [email protected] sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
