[Beowulf] A sea of wimpy cores

[Greg Lindahl]

On Fri, Sep 17, 2010 at 10:53:11AM -0400, Lawrence Stewart wrote:

> One of the other ideas at SiCortex was that a slow core wouldn't
> affect application performance of codes that were actually limited by
> the memory system.  We noticed many codes running at 1 - 5% of peak
> performance, spending the rest of their time burning a lot of power
> waiting for the memory.  I think this argument has yet to be tested,
> because the first generation SC machines didn't actually have a very
> good memory system.

This is a somewhat well-studied thing.

* Blue Gene has a node with a slow cpu, and a better memory system
than you guys had. There are probably some published studies.

* On x86, it's not hard to slow down the memory system by reducing the
# of channels, putting in slow memory, or adding more devices such
that the bus slows down. And sometimes it's possible to get the same
cpu with ddr2 and ddr3.

* There have been a ton of academic papers over the years exploring
how memory bound various codes are.

-- greg