[Info-vax] BOINC for VMS

[John Wallace]

On Mar 12, 12:48 am, Neil Rieck <n.ri... at sympatico.ca> wrote:
> On Sunday, March 11, 2012 6:32:54 AM UTC-4, Michael Kraemer wrote:
> > Hans Vlems schrieb:
>
> > > I agree that SETI and BOINC can do a lot for marketing, especially
> > > when hobbyist systems can compete too.
> > > But we know what happened to Alpha. Even if VMS was ported to Intel's
> > > x64 platform I and given proper FP support I wouldn't try to join
> > > BOINC.
>
> > Yep. I wonder if it really makes sense even on x86
> > (apart from being a funny hobby, of course).
> > The numbers one can find in the net (500k PCs, 5 Petaflops)
> > aren't really impressive when compared with today's supercomputers,
> > each of which has approx this performance level.
> > These beasts, when dedicated to stuff like protein folding,
> > are more economical and certainly more ecological than a bunch
> > of commodity PCs.
> > Today, Watts per megaflop is at least as important as purchase
> > price of hardware.
>
> I think it depends on the setup. For example, we all know that a clever programmer could play a tune with a CPU, but a special purpose device, like a sound card, is so much more efficient. Likewise, anyone who has ever attempted signal analysis with a CPU knows DSP/FFT is possible, it's just that it is much easier to do it on a CPU with MMX/SSE instructions. (SIMD technology would never have been allowed on a true RISC chip like the Alpha).
>
> But a graphics card like ATI's HD5870 is really a special purpose piece of hardware with 1600 streaming processors along with its own dedicated memory which is usually larger than the host platform.
>
> http://www3.sympatico.ca/n.rieck/docs/folding_at_home.html#ATI
>
> It is in these situations that commodity PCs really shine when doing distributed science.
>
> Neil Rieck
> Kitchener / Waterloo / Cambridge,
> Ontario, Canada.http://www3.sympatico.ca/n.rieck/


In addition to what David Froble wrote a few hours ago...

Perhaps you're not aware (publicity was never DEC's forte) but Alpha
had SIMD eventually. The Motion Video Instructions (MVI) were first
introduced in 21164PC, and found in all 21264 and later.

I don't know about x86 SIMD so can't compare the two but the DEC/
Compaq written MVI writeup at
http://www.alphalinux.org/docs/MVI-full.html
appears to be comprehensive.

One of the more unusual things that early Alphas could be used for,
even before MVI, was indeed DSP-type stuff, especially when
flexibility (of application and environment) was as important as raw
performance. DSP chips are great for performance at specific DSP
applications, but not always the world's easiest or most flexible
tools to design and code for.

There were specialist math libraries and tools available for Alpha to
simplify design and coding of DSP-type applications e.g. DEC's own
DXML, the Kuch and Associates (KAP) preprocessor for Fortran and C
available through DEC, and third party stuff like 3L's Parallel C. Or
if you had clue of the right kind, DIY.

These days there are different ways of achieving the same kind of end
result, e.g. the delegating the number crunching to a commodity
graphics chip which you mention.

Back in the day, I worked with a couple of well known names using
Alpha (21064 initially) for completely unrelated kinds of audio DSP
application, and another couple also in unrelated fields looking at
Alpha for flexible signal processing for phased array antennae.

By that stage DEC/CPQ pricing for this kind of product was
sufficiently un-outrageous for Alpha to compete technically in some
cases with both DSP and x86. Didn't last long though, after the top
level political decision was made that Alpha was on the way out
because IA64 was going to be the "industry standard 64 bit".