> I was discussing this with a friend of mine last night, and his theory was that
> the G5 is using SIMD, which the Phenom II doesn't seem to have.  I suppose that
> -could- make a difference with number crunching...
>
The 1090T supports SSE1-SSE3 SIMD extensions, according to cpu-world, however I
can't tell if it actually uses SIMD technology.

According to Everymac, the 970MP G5 processor uses SIMD and supports up to 12
simultaneous out-of-order instructions.  If it is, indeed, doing four
calculations per instruction, that might account for the difference.

looking at the Makefile for my AMD install, the -march option is set to "native"
which resulted in the build being optimized for X86_64-unknown-linux-gnu,
whereas the G5 install is optimized specifically for powerpc64-unkonwn-linux-gnu
| PowerPC G5.

Regards,
A.D.B.

Post a reply to this message

ext
"Anthony D. Baye" <Sha### [at] spamnomorehotmailcom> wrote:
> > I was discussing this with a friend of mine last night, and his theory was that
> > the G5 is using SIMD, which the Phenom II doesn't seem to have.  I suppose that
> > -could- make a difference with number crunching...
> >
> The 1090T supports SSE1-SSE3 SIMD extensions, according to cpu-world, however I
> can't tell if it actually uses SIMD technology.
>
> According to Everymac, the 970MP G5 processor uses SIMD and supports up to 12
> simultaneous out-of-order instructions.  If it is, indeed, doing four
> calculations per instruction, that might account for the difference.
>
> looking at the Makefile for my AMD install, the -march option is set to "native"
> which resulted in the build being optimized for X86_64-unknown-linux-gnu,
> whereas the G5 install is optimized specifically for powerpc64-unkonwn-linux-gnu
> | PowerPC G5.
>
> Regards,
> A.D.B.

Povray only uses SIMD registers and instructions as scalar entities (so on x86
processors, SSE2 is faster than x87).

The SIMD unit in the 970 is single precision only, so Povray will not use it.
Povray uses double precision floats. So on the 970, the regular FP unit is used.

Try this on both computers:

1) copy the benchmark.pov file to your home directory.
2) run the following command:
../povray +ibenchmark -w512 -h512 +a0.3 +v -d +f -x

This makes sure of the render size. Then compare the output images produced by
both computers.

Post a reply to this message

>
> 1) copy the benchmark.pov file to your home directory.
> 2) run the following command:
> ../povray +ibenchmark -w512 -h512 +a0.3 +v -d +f -x
>
>

D'oh! Command mangled by this website. The proper command is:

povray +ibenchmark -w512 -h512 +a0.3 +v -d +f -x

Post a reply to this message

Am 04.07.2014 21:58, schrieb Anthony D. Baye:
> "jhu" <nomail@nomail> wrote:
>> I've given this more thought. The most likely explanation is that the Mac is
>> rendering the benchmark scene at a lower resolution than it should be. The
>> default resolution for Povray 3.7 benchmark is 512x512. For Povray 3.6 it was
>> 384x384. So go check benchmark.ini and see what the height and width are set to.
>
> I'm not using 3.6 on either machine; the install is a fresh build from the git
> repository.

Which branch? The newest master, or the stable?

Jerome and I have spent the last days wrestling with Linux build 
problems in the current master; and although the most up-to-date version 
seems to build now with icpc, g++ and clang, the g++ build is 
dysfunctional and doesn't render the benchmark scene properly (and also 
too fast).

Post a reply to this message

clipka <ano### [at] anonymousorg> wrote:
> Am 04.07.2014 21:58, schrieb Anthony D. Baye:
> > "jhu" <nomail@nomail> wrote:
> >> I've given this more thought. The most likely explanation is that the Mac is
> >> rendering the benchmark scene at a lower resolution than it should be. The
> >> default resolution for Povray 3.7 benchmark is 512x512. For Povray 3.6 it was
> >> 384x384. So go check benchmark.ini and see what the height and width are set to.
> >
> > I'm not using 3.6 on either machine; the install is a fresh build from the git
> > repository.
>
> Which branch? The newest master, or the stable?
>
> Jerome and I have spent the last days wrestling with Linux build
> problems in the current master; and although the most up-to-date version
> seems to build now with icpc, g++ and clang, the g++ build is
> dysfunctional and doesn't render the benchmark scene properly (and also
> too fast).

I used the instructions posted by James Holsenback in :
http://news.povray.org/527d4309%241%40news.povray.org

git pull origin master

Regards,
A.D.B.

Post a reply to this message

"Anthony D. Baye" <Sha### [at] spamnomorehotmailcom> wrote:

>
> I used the instructions posted by James Holsenback in :
> http://news.povray.org/527d4309%241%40news.povray.org
>
> git pull origin master
>
> Regards,
> A.D.B.

You should really use the stable version for now:
https://github.com/POV-Ray/povray/archive/3.7-stable.zip

Try that one on the Mac and see what happens.

Post a reply to this message

"jhu" <nomail@nomail> wrote:
> >
> > 1) copy the benchmark.pov file to your home directory.
> > 2) run the following command:
> > ../povray +ibenchmark -w512 -h512 +a0.3 +v -d +f -x
> >
> >
>
> D'oh! Command mangled by this website. The proper command is:
>
> povray +ibenchmark -w512 -h512 +a0.3 +v -d +f -x

something seriously screwy with the mac output.  Like half the image is black,
which I guess accounts for the speed.

So the question now is: Why?

Regards,
A.D.B

Post a reply to this message

"Anthony D. Baye" <Sha### [at] spamnomorehotmailcom> wrote:

> something seriously screwy with the mac output.  Like half the image is black,
> which I guess accounts for the speed.
>
> So the question now is: Why?
>
> Regards,
> A.D.B

I don't know. I have Debian 7 on a Power Mac G4. Compiled the stable 3.7 just
fine. It's as slow as expected when rendering the benchmark scene. Try compiling
using the stable source I linked to. Otherwise I can send you the binary I have.

Post a reply to this message

Another thing to check: Try rendering some other scene (with the exact
same settings) to see how the difference in time goes.

-- 
                                                          - Warp

Post a reply to this message

Warp <war### [at] tagpovrayorg> wrote:
> Another thing to check: Try rendering some other scene (with the exact
> same settings) to see how the difference in time goes.
>
> --
>                                                           - Warp


Well, the settings aren't the same, but I did render my lotus gas lamp and it
looks just fine... Though, as expected, it does take much longer.

I think there might be a problem with the calculation of mesh objects or mesh2
objects in my build because looking at the output image from rendering the
benchmark, there seem to be a lot of missing surfaces.

Regards,
A.D.B.

Post a reply to this message