[Beowulf] Parallel programming for Xeon Phis
e.scott.atchley at gmail.com
Wed Aug 24 14:56:27 PDT 2016
On Wed, Aug 24, 2016 at 4:54 PM, Greg Lindahl <lindahl at pbm.com> wrote:
> On Wed, Aug 24, 2016 at 04:44:03PM +0000, John Hearns wrote:
> > OK, I guess that the state of the art for a FORTRAN Compiler in the
> > 60s is pitiful compared to the sophisticated compilers we have
> > today.
> Actually, Fortran was designed to be optimized, and IBM's first
> compiler was an optimizing compiler. A compiler/language co-design, if
> you will. It wasn't very sophisticated compared to today's compilers,
> but it was extremely sophisticated for its era.
> As for Doug's comments about Summit and Sierra being co-designs, I
> think that's great. InfiniPath's design was focused on building the
> smallest possible chip that did what MPI needed. I've always wondered
> what would have happened to Myricom had they redesigned their chip to
> do exactly what their MX software interface needed.
> -- greg
Given its research origins, the Myricom designs remained full programmable,
which meant the fast path was not in hardware.
When I left to go to ORNL, work was being done to put the fast path in
hardware while retaining the ability for alternate, fully programmable
paths for custom workloads (e.g. video streaming, trading).
That IP was sold before the company was bought by CSPI and much of the
staff went with the IP to its current home.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Beowulf