[Beowulf] El Reg: AMD reveals potent parallel processing breakthrough
Lux, Jim (337C)
james.p.lux at jpl.nasa.gov
Mon May 13 06:19:10 PDT 2013
On 5/12/13 10:36 PM, "Vincent Diepeveen" <diep at xs4all.nl> wrote:
>On May 12, 2013, at 8:18 PM, Geoffrey Jacobs wrote:
>> On 05/12/2013 12:25 PM, Vincent Diepeveen wrote:
>>> I sense clearly that you keep overlooking the BIGGEST problem.
>>> Namely that the guy at home has a car and he feels he can drive
>>> it. In
>>> fact he can.
>>> And that you are going to teach the guy how to drive a moped.
>>> Are you realizing what i try to make clear all the time?
>> Bad analogy. A better one is:
>It is exactly a good analogy, because a raspberry-pi cluster is going
>to be a LOT SLOWER
>than the cpu + gpu that the student has at home.
But if the student wants to learn how to do distributed computing with
different topologies, speed isn't important. Whether I invert a 5x5
matrix in a microsecond or a millisecond or a second makes no difference,
as long as I can measure it. And that I can instrument my code to
understand the relative impacts of things like communication speed vs
computation speed, and how that changes as the scaling changes.
The decision about whether to do this in simulation with a bunch of VMs or
with discrete hardware is more an aesthetic one. I happen to believe that
having lots of boxes on the table top is more interesting, but then, I'm
the kind of guy who knows which end of the soldering iron is hot. My
wife, on the other hand, says that if you're a software person, you
shouldn't care what the underlying hardware is, because at work, you'll
never get to see or touch it anyway. (one might say that this reliance on
abstracting away the hardware is the root of all HPC or CS evil, but
that's another discussion).
There is a rich panoply of pedagogical approaches. You have your preferred
ones, I have mine, I'm pretty sure everyone on the list has opinions of
what they think. All I am doing is proposing a way that *I* would find
attractive if *I* were a student. I'm probably atypical.
>Now the less talented students might follow your course, get a degree
>and then start a PHD - which is one of the problems over here in
>The Netherlands and if i look at the facts it's similar problem in USA.
And how is this a problem? We've found someone who's actually interested
in computer engineering. Better than someone choosing a major based on
statistical properties of integrated lifetime earnings as a function of
major and institution. Whether they make the grade in their PhD program
is determined by a lot of other things, but philosophically, I'd rather
have the candidates have a passion, rather than a spreadsheet and a model.
(My wife, the MBA managing software people, may have different
(but, then, I *am* a sort of child of the sixties who grew up with the
whole eudaemonic "do what feels right", "follow your passion" thing, etc.;
Perhaps todays students, in general, have a more green eyeshade view. Mine
don't, as much, but certainly, some of their friends do)
More information about the Beowulf