In article <jer### [at] cerebusacusdedu> , jer### [at] acusdedu (Jerry)
wrote:

>Is there any way to tell the operating system in a dual-processor
>environment to use both processors in floating-point operations? Does

There is no difference for processor between floating point and integer instructions,
the thread with *all* its instructions will be exectuted on the same processor. For
todays processors (with built-in floating point units) there is *no* difference
between floating point and integer or any other instructions.

>processor-switching occur fast enough for small operations like this to
>speed up otherwise non-dual-supporting applications?

I suppose this is a typo and you wanted to write process-switching :-)
Well, if there are two tasks each will run on a different processor (also this might
depend on lots of there things).  As there are running lots of other tasks as well
(like the Windows Explorer (the desktop), network services etc) and those are always
there even on a single processor computer you will get a noticeable speed increase,
if you have enough memory so no constant swapping is required you will see more than
50% speed increase when running two POV-Ray version at the same time. You can
estimate the speed increase this way: Take one of the demo images that come with
POV-Ray and that take a few minutes to render (the longer the rendering takes the
more precise the measurement). Now render it when only one POV-Ray is running and get
the render time (this is time 1).  After that start a second POV-Ray and load the
same scene file. Now prepare both POV-Rays for a render start. Start the first one,
then (hurry!) switch to the second POV-Ray and start it as well. Wait after both have
finished and get the render times (these are times 2 and 3). How much faster was the
rendering 1 compared to the _average_ of times 2 and 3. If rendering 1 took only half
the time of the average only one processor was used, if the average is near (perhaps
20% difference) time 1 you can be (nearly) sure both processors were used.


    Thorsten

Post a reply to this message

Thorsten Froehlich wrote:
> 
> In article <jer### [at] cerebusacusdedu> , jer### [at] acusdedu (Jerry)
> wrote:
> 
> >Is there any way to tell the operating system in a dual-processor
> >environment to use both processors in floating-point operations? Does
> 
> There is no difference for processor between floating point and integer
instructions,
> the thread with *all* its instructions will be exectuted on the same processor. For
> todays processors (with built-in floating point units) there is *no* difference
> between floating point and integer or any other instructions.

I don't get what you mean.  There are separate machine instructions and
registers for floating point.  How is ther no difference.  Of course
I've never programed for an Intel.


> 
> >processor-switching occur fast enough for small operations like this to
> >speed up otherwise non-dual-supporting applications?
> 
> I suppose this is a typo and you wanted to write process-switching :-)
> Well, if there are two tasks each will run on a different processor (also this might
> depend on lots of there things).  As there are running lots of other tasks as well
> (like the Windows Explorer (the desktop), network services etc) and those are always
> there even on a single processor computer you will get a noticeable speed increase,
> if you have enough memory so no constant swapping is required you will see more than
> 50% speed increase when running two POV-Ray version at the same time. You can
> estimate the speed increase this way: Take one of the demo images that come with
> POV-Ray and that take a few minutes to render (the longer the rendering takes the
> more precise the measurement). Now render it when only one POV-Ray is running and
get
> the render time (this is time 1).  After that start a second POV-Ray and load the
> same scene file. Now prepare both POV-Rays for a render start. Start the first one,
> then (hurry!) switch to the second POV-Ray and start it as well. Wait after both
have
> finished and get the render times (these are times 2 and 3). How much faster was the
> rendering 1 compared to the _average_ of times 2 and 3. If rendering 1 took only
half
> the time of the average only one processor was used, if the average is near (perhaps
> 20% difference) time 1 you can be (nearly) sure both processors were used.
> 
>     Thorsten

Post a reply to this message