[rescue] SMP on intel wasteful?

[Francisco Javier Mesa-Martinez]

On Wed, 26 Jun 2002, Bjorn Ramqvist wrote:

> Is Cray phasing out their T3-line?
> To me, that sounds like a big pile of bull-####, but it makes sense in
> one way; it's old.


> Oh yes, put faster processors in it, but that would defeat the overall
> machine in the end. That machine (or those machines) were designed for a
> specific purpose with specific processors and all that. No way they're
> getting scalable juice out of that by just putting some anabolics here
> and speed up there.
> 
> > The truth is that discussions like these are moot. Vector is hardly used
> > anymore, and when it is usually specialized SIMD approaches are used in
> > that field. Granted each approach has its niche, clusters are better for
> > non coupled data problems, whereas vector/SIMD machines are better for
> > data parallel problems. Once you have specified your problem, then by al
> > means chose the RIGHT tool to solve it, And let's stop this pissing
> > contest, please.
> 
> What does the T3 have in common with vector machines? Nothing.
> It was designed by a company that makes both vector and parallel
> machines. Other than that?

Where the heck did I discuss the T3 in that paragraph?

> And, from what I know, you don't do parallel computations on a vector
> machine, or the other way around. That kind of defeats its purpose...

Vector machines need data parallelism in order to be efficient, and most
modern vector machines are SMP in one form or another. Vector
optimizations usually work by finding parallelism at the loop level, there
are many kinds of parallelism you know, not only at the instruction level.
So if you do not do parallel computations on a vector machines, i.e. you
are running plain scalar code.... then you are just using the scalar units
on the machine (which are not very optimized in most vector machines). So
that is pretty much a waste.