DSP 'Beowulf'

Hello one and all,

I have been suffering from the desire to do some heavy math.
Yes, madness comes in many forms..

For this particular project, I would like to put together
a Beowulf of sorts (except that it will not be running the
Linux (Beowulf-in-a-Box) code.

It has been a few years since I have been involved with the
hardware side of things. So, I have two questions:

 1. Are DSP chips still the best way to go for heavy numerical
    work? Or, should I go with a stack of MMX mainboards...

 2. If DSP is still The Way, are there any older consumer products
    that contain a robust DSP that can be programmed? That is, 40
    identical sound cards, or modems, come within budget - as long
    as I can get code to the processor.

    40 ADI eval boards are not within the budget.

Best regards,
Jerry Parks

Jerry Parks wrote:
> 
> Hello one and all,
> 
> I have been suffering from the desire to do some heavy math.
> Yes, madness comes in many forms..
> 
> For this particular project, I would like to put together
> a Beowulf of sorts (except that it will not be running the
> Linux (Beowulf-in-a-Box) code.
> 
> It has been a few years since I have been involved with the
> hardware side of things. So, I have two questions:
> 
>  1. Are DSP chips still the best way to go for heavy numerical
>     work? Or, should I go with a stack of MMX mainboards...
> 
>  2. If DSP is still The Way, are there any older consumer products
>     that contain a robust DSP that can be programmed? That is, 40
>     identical sound cards, or modems, come within budget - as long
>     as I can get code to the processor.
> 
>     40 ADI eval boards are not within the budget.
> 
> Best regards,
> Jerry Parks

Watch out for Grendel-in-a-cave. If you don't need low power and
portability, you will probably be better off with MMX.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

"Jerry Avins" <jya@ieee.org> wrote in message
news:3F5910F6.6A79DF3A@ieee.org...
> Jerry Parks wrote:
> >
> > Hello one and all,
> >
> > I have been suffering from the desire to do some heavy math.
> > Yes, madness comes in many forms..
> >
> > For this particular project, I would like to put together
> > a Beowulf of sorts (except that it will not be running the
> > Linux (Beowulf-in-a-Box) code.
> >
> > It has been a few years since I have been involved with the
> > hardware side of things. So, I have two questions:
> >
> >  1. Are DSP chips still the best way to go for heavy numerical
> >     work? Or, should I go with a stack of MMX mainboards...
> >
> >  2. If DSP is still The Way, are there any older consumer products
> >     that contain a robust DSP that can be programmed? That is, 40
> >     identical sound cards, or modems, come within budget - as long
> >     as I can get code to the processor.
> >
> >     40 ADI eval boards are not within the budget.
> >
> > Best regards,
> > Jerry Parks
>
> Watch out for Grendel-in-a-cave. If you don't need low power and
> portability, you will probably be better off with MMX.

I haven't actually researched it but I'll bet you find that the PC
mainboards improve in speed and capacity faster than the DSP chips.
Accordingly, they will take market share in some situations.  It wasn't that
long ago that 130MHz Pentium machines were occasionally taking sales away
from top of the line DSP cards - because developers discovered they could do
the job.  The way I look at it is that existing applications are continually
being embraced by PCs as the PC performance increases.  What almost can be
done today with a PC will be doable next year, etc.  If you assume that any
one defined application demands a fixed amount of processing power then some
such defined applications will be newly reachable each year by PCs.

So, Jerry's comment about low power and portability are right on the mark!
In addition, relatively massive processing needs will still be served by DSP
machines of one sort or another - simply because they've been designed to do
the job.  Yeah, maybe some day we'll see the end of this being the case but
I'm not going to hold my breath!

If you assume that the appetite for processing power grows as experience
with real increased processing power grows - just ask a pragmatist - then
you might imagine that the "unreachable" sement of the market (for PCs)
continues to grow about as fast as the "newly reachable" segment grows.  No
numbers here, just a philosophical observation.

When it comes to selecting an architecture, the application will determine.
So your question is pretty hard to answer because the application isn't
revealed.  For example, if processing is more serial than parallel and
latency needs to be small then will a parallel architecture on a serial
application with rather large amounts of data transfer be appropriate?  In
some cases a parallel architecture doesn't help at all!

BTW: "DSP chips" don't make an architecture - they might only imply an
architecture.
And there's the question of software overhead: operating system, etc.

It's quite possible that someone has tackled the same problem you have in
mind.  What is it?

Fred

Jerry Parks wrote:

>   Nothing too exotic - just taking the product of large primes.
>   It is unfortunate that there are so many of them...  :)
> 
check out the professionals: http://arrakis.ncsa.uiuc.edu/ps2/cluster.php
They use playstations!  I've always wanted to go dumpster diving and just
collect a whole lot of those and rig them up for this kind of task.  Right
now, playstation 1's are available at my local game store for US$30 each,
so 10 of those will beat out most any 32 bit processor (they are 64 bit,
the playstation 2 is 128 bit).  For large integer math, these things are
a lot cheaper than any fpga, and gluing them together isn't all that hard.

The programming tools are also pretty cheap and easy to get as well.

Patience, persistence, truth,
Dr. mike

-- 
Mike Rosing
www.beastrider.com                   BeastRider, LLC
SHARC debug tools

"Jerry Parks" <j.parks@relmail.com> wrote in message =
news:p4nhlv4ebd7rmh2oi0kraec5gdeh9pu3fq@4ax.com...
>=20
>=20
> Hello one and all,
>=20
> I have been suffering from the desire to do some heavy math.
> Yes, madness comes in many forms..
>=20
> For this particular project, I would like to put together
> a Beowulf of sorts (except that it will not be running the
> Linux (Beowulf-in-a-Box) code.
>=20
> It has been a few years since I have been involved with the
> hardware side of things. So, I have two questions:
>=20
>  1. Are DSP chips still the best way to go for heavy numerical
>     work? Or, should I go with a stack of MMX mainboards...
>=20
>  2. If DSP is still The Way, are there any older consumer products
>     that contain a robust DSP that can be programmed? That is, 40
>     identical sound cards, or modems, come within budget - as long
>     as I can get code to the processor.
>=20
>     40 ADI eval boards are not within the budget.
>=20
> Best regards,
> Jerry Parks
>=20

mmx ? sure you don't mean sse2.
system streaming extensions. intels improved mmx

For some workloads altivec (mmx equivalent for powerpc )
leaves sse2 for dead.

options for powerpc either IBM or  motorola.

beowulf wise maybe look at some opteron boxes
or a blade server.

Alex

Alex Gibson wrote:
> mmx ? sure you don't mean sse2.
> system streaming extensions. intels improved mmx
> 
> For some workloads altivec (mmx equivalent for powerpc )
> leaves sse2 for dead.
> 
> options for powerpc either IBM or  motorola.
> 
> beowulf wise maybe look at some opteron boxes
> or a blade server.
> 
> Alex
When using the integer part of SSE2, I have found it generally performs 
slightly worse than MMX. For floats it works somewhat better. Its all a 
pretty lousy form of DSP add on, though. Its a big struggle with any 
part of MMX/SSE/SSE2 to get any speed up with many common algorithms, 
like LMS.

Regards,
Steve

Jerry Parks wrote:

> 
> 
> Hello one and all,
> 
> I have been suffering from the desire to do some heavy math.
> Yes, madness comes in many forms..
> 
> For this particular project, I would like to put together
> a Beowulf of sorts (except that it will not be running the
> Linux (Beowulf-in-a-Box) code.
> 
> It has been a few years since I have been involved with the
> hardware side of things. So, I have two questions:
> 
>  1. Are DSP chips still the best way to go for heavy numerical
>     work? Or, should I go with a stack of MMX mainboards...
> 
>  2. If DSP is still The Way, are there any older consumer products
>     that contain a robust DSP that can be programmed? That is, 40
>     identical sound cards, or modems, come within budget - as long
>     as I can get code to the processor.
> 
>     40 ADI eval boards are not within the budget.
> 
> Best regards,
> Jerry Parks

Hi Jerry,
where come the data from, which you're going to process?
Is your concept limiting the method how you connect the processors?
Because you mention sound cards and ADI eval boards, I guess you're 
processing parallel data, aren't you?

If input values come from parallel sources, look at the SPORT of 
ADI, where it is possible to process 8 or even more channels 
through a DMA, so that it "costs" nothing on the processor side.
This DMA concept works too if you link two or more DSPs - no 
overhead at all! 
However, if '40' comes from 40 signals which you're going to 
process, you'd probably need at most 5 DSPs to receive them and not 
40. Are 5 eval boards within your reach?  

On the other hand, if I imagine a mesh of 40 Sharc DSPs with a 
minimum hardware environment, you'd probably receive an impressive 
calculation power 
at moderate price. And it would certainly fit very well into the 
Beowulf concept (sorry, if I'm wrong: I'm not too familiar with 
that :-) - at least not much programming efforts...

Another approach which comes to my mind:
FPGA have improved very much - in fact, you can even implement a 
couple of Pentium processors on one chip (probably too expensive).
Depending on what "heavy math" means, and if it's more important to 
have a pretty design or to get a problem solved,  it might be worth 
to check if an approach using FPGAs would be faster or better. 

I guess FPGAs would win the race as soon as you want to do something 
which is aside the usual and where the available processors on 
their as-is-basis don't fit. The biggest advantage of FPGAs is 
certainly that they are tailorable to whichever need.
There are eval boards available for FPGAs, too.

Just my 2c.

Bernhard

-- 
before sending to the above email-address:
replace deadspam.com by foerstergroup.de

On Fri, 5 Sep 2003 23:02:33 -0700, "Fred Marshall"
<fmarshallx@remove_the_x.acm.org> wrote:


<snip>
>When it comes to selecting an architecture, the application will determine.
>So your question is pretty hard to answer because the application isn't
>revealed.  For example, if processing is more serial than parallel and
>latency needs to be small then will a parallel architecture on a serial
>application with rather large amounts of data transfer be appropriate?  In
>some cases a parallel architecture doesn't help at all!
>
>BTW: "DSP chips" don't make an architecture - they might only imply an
>architecture.
>And there's the question of software overhead: operating system, etc.
>
>It's quite possible that someone has tackled the same problem you have in
>mind.  What is it?
>
>Fred
>


My apologies for not having gave you more details. It really isn't
numerical analysis or processing as much as it is brute force
multiplying large primes and factoring largish integers.

Jerry

On Sun, 7 Sep 2003 21:25:58 +1000, "Alex Gibson" <alxx@ihug.com.au>
wrote:

<snip>
>mmx ? sure you don't mean sse2.
>system streaming extensions. intels improved mmx

  Like I said, I'm a bit out of touch on the hardware side
  of things.

>For some workloads altivec (mmx equivalent for powerpc )
>leaves sse2 for dead.

  Yes, I just noticed Altivec shortly after my original posting.


>options for powerpc either IBM or  motorola.
>
>beowulf wise maybe look at some opteron boxes
>or a blade server.
>
>Alex


  What I would really like to see is a PPC SBC. Just the processor,
  some RAM, and a way to talk to to it...

  Jerry

On Mon, 08 Sep 2003 08:43:18 +0200, Bernhard Holzmayer
<holzmayer.bernhard@deadspam.com> wrote:


>where come the data from, which you're going to process?
>Is your concept limiting the method how you connect the processors?
>Because you mention sound cards and ADI eval boards, I guess you're 
>processing parallel data, aren't you?
>
>If input values come from parallel sources, look at the SPORT of 
>ADI, where it is possible to process 8 or even more channels 
>through a DMA, so that it "costs" nothing on the processor side.
>This DMA concept works too if you link two or more DSPs - no 
>overhead at all! 
>However, if '40' comes from 40 signals which you're going to 
>process, you'd probably need at most 5 DSPs to receive them and not 
>40. Are 5 eval boards within your reach?  

  Nothing too exotic - just taking the product of large primes.
  It is unfortunate that there are so many of them...  :)


>On the other hand, if I imagine a mesh of 40 Sharc DSPs with a 
>minimum hardware environment, you'd probably receive an impressive 
>calculation power 
>at moderate price. And it would certainly fit very well into the 
>Beowulf concept (sorry, if I'm wrong: I'm not too familiar with 
>that :-) - at least not much programming efforts...
>
>Another approach which comes to my mind:
>FPGA have improved very much - in fact, you can even implement a 
>couple of Pentium processors on one chip (probably too expensive).
>Depending on what "heavy math" means, and if it's more important to 
>have a pretty design or to get a problem solved,  it might be worth 
>to check if an approach using FPGAs would be faster or better. 
>
>I guess FPGAs would win the race as soon as you want to do something 
>which is aside the usual and where the available processors on 
>their as-is-basis don't fit. The biggest advantage of FPGAs is 
>certainly that they are tailorable to whichever need.
>There are eval boards available for FPGAs, too.
>
>Just my 2c.

  Worth much more than that, I would say.

>
>Bernhard


  Jerry