hi all,

how does the current implementation compare with the transputer port 20 
years ago? i haven't found any further information on that project. i would 
be curious, how the dependency analysis has been carried out in those days?


> What is "fast enough?" In my view it is fast enough if (a) complex
> pieces run significantly faster with 2 or more cores, and (b)
> performance shows some kind of scaling up with increasing numbers of
> cores. I don't think it's fast enough if adding cores does no good or
> even slows things down again.

one more point to this: how does the real-time behavior scale? while 
throughput is one aspect, i have the experience, that the real-time 
performance scales less well than the throughput ...

tim

-- 
tim@klingt.org
http://tim.klingt.org

Only very good and very bad programmers use goto in C



------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
_______________________________________________
Csound-devel mailing list
Csound-devel@lists.sourceforge.net

I had a copy of the transputer papers which I made for Richard Dobson
but I cannot remember where I put them, but I do not think that they
had the interlocking analysis idea, which comes from Jed Marti's
thesis (U of utah) and work we did in RAND in the 1980s.  Richard may
be able to answer the question on the comparison.

I expect that there will be a slow down if the number of threads
exceeds real cores and if the orchestra does not have instrument-level
parallelism.  At present I have an overhead problem I think

==John ffitch

Still looking for the Transputer papers

>>>>> "Tim" == Tim Blechmann  writes:

 Tim> hi all,

 Tim> how does the current implementation compare with the transputer port 20 
 Tim> years ago? i haven't found any further information on that project. i would 
 Tim> be curious, how the dependency analysis has been carried out in those days?


 >> What is "fast enough?" In my view it is fast enough if (a) complex
 >> pieces run significantly faster with 2 or more cores, and (b)
 >> performance shows some kind of scaling up with increasing numbers of
 >> cores. I don't think it's fast enough if adding cores does no good or
 >> even slows things down again.

 Tim> one more point to this: how does the real-time behavior scale? while 
 Tim> throughput is one aspect, i have the experience, that the real-time 
 Tim> performance scales less well than the throughput ...

 Tim> tim



------------------------------------------------------------------------------
The Palm PDK Hot Apps Program offers developers who use the
Plug-In Development Kit to bring their C/C++ apps to Palm for a share 
of $1 Million in cash or HP Products. Visit us here for more details:
http://ad.doubleclick.net/clk;226879339;13503038;l?
http://clk.atdmt.com/CRS/go/247765532/direct/01/
_______________________________________________
Csound-devel mailing list
Csound-devel@lists.sourceforge.net

I am interested in both real-time performance and in throughput, but
throughput is more important in the end as it is the final limit on
what the composer can realize.

I think it is instructive to review the history of 3-dimensional
computer graphics rendering in the movies. Because each frame can be
rendered independently the job is inherently parallel and can be
scaled up quite bit, so we now  have "render farms" that can compute
an entire film of truly 3-dimensional, high-resolution, ray-traced and
textured computer animation. The amount of computer power in these
things is staggering.

My point is simply that artists will find ways to use up all computer
power on offer. I don't think this will ever stop.

Regards,
Mike

On Tue, Jul 27, 2010 at 12:41 PM, john ffitch  wrote:
> I had a copy of the transputer papers which I made for Richard Dobson
> but I cannot remember where I put them, but I do not think that they
> had the interlocking analysis idea, which comes from Jed Marti's
> thesis (U of utah) and work we did in RAND in the 1980s.  Richard may
> be able to answer the question on the comparison.
>
> I expect that there will be a slow down if the number of threads
> exceeds real cores and if the orchestra does not have instrument-level
> parallelism.  At present I have an overhead problem I think
>
> ==John ffitch
>
> Still looking for the Transputer papers
>
>>>>>> "Tim" == Tim Blechmann  writes:
>
>  Tim> hi all,
>
>  Tim> how does the current implementation compare with the transputer port 20
>  Tim> years ago? i haven't found any further information on that project. i would
>  Tim> be curious, how the dependency analysis has been carried out in those days?
>
>
>  >> What is "fast enough?" In my view it is fast enough if (a) complex
>  >> pieces run significantly faster with 2 or more cores, and (b)
>  >> performance shows some kind of scaling up with increasing numbers of
>  >> cores. I don't think it's fast enough if adding cores does no good or
>  >> even slows things down again.
>
>  Tim> one more point to this: how does the real-time behavior scale? while
>  Tim> throughput is one aspect, i have the experience, that the real-time
>  Tim> performance scales less well than the throughput ...
>
>  Tim> tim
>
>
>
> ------------------------------------------------------------------------------
> The Palm PDK Hot Apps Program offers developers who use the
> Plug-In Development Kit to bring their C/C++ apps to Palm for a share
> of $1 Million in cash or HP Products. Visit us here for more details:
> http://ad.doubleclick.net/clk;226879339;13503038;l?
> http://clk.atdmt.com/CRS/go/247765532/direct/01/
> _______________________________________________
> Csound-devel mailing list
> Csound-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/csound-devel
>



-- 
Michael Gogins
Irreducible Productions
http://www.michael-gogins.com
Michael dot Gogins at gmail dot com

------------------------------------------------------------------------------
The Palm PDK Hot Apps Program offers developers who use the
Plug-In Development Kit to bring their C/C++ apps to Palm for a share 
of $1 Million in cash or HP Products. Visit us here for more details:
http://ad.doubleclick.net/clk;226879339;13503038;l?
http://clk.atdmt.com/CRS/go/247765532/direct/01/
_______________________________________________
Csound-devel mailing list
Csound-devel@lists.sourceforge.net