Unfortunatly, 1 CPU + 1 CPU is not equal to 2 CPUs ... :)

When you have a 2 CPUs machine, the processors share the same front side
bus, and the same memory modules. A direct concequence is that 2 instances
running in the same time, will not share the same data set. The concequence
will be that you will double the memory traffic. This traffic is extremely
slow compare to the 2 or 3 GHz of the processors, and you will probably see
around 10% to 25% speed improvement with a seconds instance running
concurently, with 2 physical processors.

What you really want to do is the have 2 treads running using the same
"source data" in reading mode only (No L2 cache write combine between the 2
processors), both thread have their own local variables, and 2 differents
input.

The dream case, is to have a commun L2 cache, with 2 cores (or 2
Hyperthreading logical CPUs) and get the source data inside the L2 Cache (or
L3) and try to maximize the locality of the source data by rendering the
line n on the CPU n, and line n+1 on CPU(n+1) etc ...
Doing so, you will minimize the 1st L2 cache miss (statistically, 2 close
pixels share the same 3D collisions, at least the 1st one on the same
object).

I am testing those ideas on Xeon with Hyperthreading technology, and I am
seeing good performance improvement.

Of course, I did not test all the solutions, and an algorythm solution may
be better than the "brut force" solution I am offering.

I will be releasing a sample code doing so soon, based on PovRay 3.5

If you want to learn about Hyperthreading :
http://cedar.intel.com/media/training/hyper_threading_intro/tutorial/index.h
tm (Click on next at the bottom)
Cool training ;)


Francois

--------------------------------------------------------------
Francois Piednoel
Senior Performance Analyst
Pentium 4, Prescott, Tejas.
Intel Corporation




"Lance Birch" <-> wrote in message news:3d36a55f@news.povray.org...
> OK, as was suggested, running multiple instances simultaneously to split
the
> work load is the way to go right now.  What I'd like to know (and it seems
a
> really silly question), is how to force POV-Ray to run two instances?
>
> I have a rendering going along right now that's only at 5 PPS (what can I
> say, the faster the computer, the more complex my scenes get!) and I'd
like
> to have it at 10 by running two instances with the affinity fixed to CPU0
> for one and CPU1 for the second, however, when I try to run a second
> instance, the focus just changes to the existing instance as though it's
> designed not to allow multiple instances.
>
> What am I doing wrong here... is there a command line switch to tell it to
> force a second instance?  I had a look in the help files but I couldn't
find
> anything.
>
> Regards,
>
> Lance.
>
> thezone.firewave.com.au
> www.firewave.com.au
>
>

Post a reply to this message