Radiance timings on PowerMac

Hi Mark,

I ran your test on my 1.42 GHz PowerMac running OS X 10.2.6 using Radiance 3.5 compiled with gcc 3.1 -O2, and got the following results:

oconv scene.rad
26.100u 2.050s 0:23.08 121.9% 0+0k 0+2io 0pf+0w
rpict -t 30 -vf v1.vf -ab 1 -x 1000 -y 1000 scene.oct
rpict: 34274852 rays, 100.00% after 0.144u 0.000s 0.152r hours on anyhere.local.
512.010u 6.820s 9:06.75 94.8% 0+0k 9+4io 0pf+0w
pfilt -x /2 -y /2 scene.pic
0.560u 0.010s 0:00.59 96.6% 0+0k 0+0io 0pf+0w

Nice webpage!
-Greg

Hi Mark,

Yes, my PowerMac is a dual-processor, so the system was taking advantage of the second processor on the piped command output used by oconv, apparently. Two rpict processes would finish in the same time, and both processors can be used simultaneously by programs such as ranimate, ranimove, rpiece, rholo, and some of the scripts (like falsecolor).

-Greg

···

From: Mark Stock <[email protected]>
Date: Wed Jun 4, 2003 11:01:54 AM US/Pacific
To: Greg Ward <[email protected]>
Subject: Re: Radiance timings on PowerMac

Is that a dual-processor machine, Greg? The only place I've
ever seen user times greater than wall clock times is for
oconv on a dual-proc machine.

Mark

On Wed, 4 Jun 2003, Greg Ward wrote:

Hi Mark,

I ran your test on my 1.42 GHz PowerMac running OS X 10.2.6 using
Radiance 3.5 compiled with gcc 3.1 -O2, and got the following results:

oconv scene.rad
26.100u 2.050s 0:23.08 121.9% 0+0k 0+2io 0pf+0w
rpict -t 30 -vf v1.vf -ab 1 -x 1000 -y 1000 scene.oct
rpict: 34274852 rays, 100.00% after 0.144u 0.000s 0.152r hours on
anyhere.local.
512.010u 6.820s 9:06.75 94.8% 0+0k 9+4io 0pf+0w
pfilt -x /2 -y /2 scene.pic
0.560u 0.010s 0:00.59 96.6% 0+0k 0+0io 0pf+0w

Nice webpage!
-Greg

Greg,

Yes, I've noticed that. I'm finally sitting on a dual-processor
box here at work, though I have insufficient RAM here to render
the kinds of scenes that I want. The image for SIGGRAPH this
year will probably take up 1.5 GB of RAM once the ambient cache
becomes fully populated. I'm shooting for a final resolution of
28800x14400 pixels.

A year or two ago, someone submitted a speed test for the
same scene, Paul's bench2, using a dual-processor box and
rad/rpiece. See note 8 on my bechmark page. The author found
a 1.6 speedup for dual processors using 20-25 divisions in
rad/rpiece.

Is there no newer gcc for OSX than 3.1?

Mark

···

On Wed, 4 Jun 2003, Greg Ward wrote:

Yes, my PowerMac is a dual-processor, so the system was taking
advantage of the second processor on the piped command output used by
oconv, apparently. Two rpict processes would finish in the same time,
and both processors can be used simultaneously by programs such as
ranimate, ranimove, rpiece, rholo, and some of the scripts (like
falsecolor).

-Greg

> From: Mark Stock <[email protected]>
> Date: Wed Jun 4, 2003 11:01:54 AM US/Pacific
> To: Greg Ward <[email protected]>
> Subject: Re: Radiance timings on PowerMac
>
> Is that a dual-processor machine, Greg? The only place I've
> ever seen user times greater than wall clock times is for
> oconv on a dual-proc machine.
>
> Mark
>
> On Wed, 4 Jun 2003, Greg Ward wrote:
>
>> Hi Mark,
>>
>> I ran your test on my 1.42 GHz PowerMac running OS X 10.2.6 using
>> Radiance 3.5 compiled with gcc 3.1 -O2, and got the following results:
>>
>> oconv scene.rad
>> 26.100u 2.050s 0:23.08 121.9% 0+0k 0+2io 0pf+0w
>> rpict -t 30 -vf v1.vf -ab 1 -x 1000 -y 1000 scene.oct
>> rpict: 34274852 rays, 100.00% after 0.144u 0.000s 0.152r hours on
>> anyhere.local.
>> 512.010u 6.820s 9:06.75 94.8% 0+0k 9+4io 0pf+0w
>> pfilt -x /2 -y /2 scene.pic
>> 0.560u 0.010s 0:00.59 96.6% 0+0k 0+0io 0pf+0w
>>
>> Nice webpage!
>> -Greg