...has been broken!
That's right, people. One of the beloved render farm machines at my office just turned in a sub-4K second timing on the Mark Stock Radiance benchmark scene. SSSssssssss!
What's interesting to note is that there are subtle differences in cpu speed, even among the same cpu model. I assume this is due to different voltage settings on the motherboards, but I'm not positive. I also wonder how much of the speedup is attributable to the use of the new official release of Radiance 3.8. Another interesting thing is that three of four benchmarks had the same exact number of rays traced. I find this a little odd.
I was hoping Greg and maybe Ferdinand (builder and maintainer of my office's render farm systems) could weigh in on this. I have a couple other benchmark timings and stats to add to the site (http://mark.technolope.org/pages/rad_bench.html) as well, I was hoping others have installed the official 3.8 and were doing similar testing.
Happy benchmarking.
- Rob
regarding [Radiance-general] The sub-4K second benchmark barrier...:
...has been broken!
That's right, people. One of the beloved render farm machines at my
office just turned in a sub-4K second timing on the Mark Stock Radiance
benchmark scene. SSSssssssss!
I did 3857.2 last week when testing my 3.8 Blastwave packages. That
with "only" a 2193 Mhz chip. Must send the results off.
Solaris 10. Sun Studio 11 compiler. Opteron@2193MHz.
Generic package with isaexec arch parts. This still shows 3.7.2 but
see:
http://ww.blastwave.org/pacakges/radiance
I also
wonder how much of the speedup is attributable to the use of the new
official release of Radiance 3.8.
3.8 is slightly faster but I didn't spend long comparing. I did a
test because I wanted to understand if the improvement on my previous
score was due to 3.8 or my newer compiler. The answer is the Sun
Studio 11 compiler is much better.
James.
···
On 01/11/06, 17:40:02, Rob Guglielmetti <[email protected]> wrote
I haven't benchmarked the "official 3.8" binaries on the G5, G4, or Intel as yet, but I should get around to that and post the results. I don't know what I might have done, offhand, to improve the performance over last year's entries.
-Greg
···
From: Rob Guglielmetti <[email protected]>
Date: November 1, 2006 9:40:02 AM PST
...has been broken!
That's right, people. One of the beloved render farm machines at my office just turned in a sub-4K second timing on the Mark Stock Radiance benchmark scene. SSSssssssss!
What's interesting to note is that there are subtle differences in cpu speed, even among the same cpu model. I assume this is due to different voltage settings on the motherboards, but I'm not positive. I also wonder how much of the speedup is attributable to the use of the new official release of Radiance 3.8. Another interesting thing is that three of four benchmarks had the same exact number of rays traced. I find this a little odd.
I was hoping Greg and maybe Ferdinand (builder and maintainer of my office's render farm systems) could weigh in on this. I have a couple other benchmark timings and stats to add to the site (http://mark.technolope.org/pages/rad_bench.html\) as well, I was hoping others have installed the official 3.8 and were doing similar testing.
Happy benchmarking.
- Rob
On 01/11/06, 18:04:12, James Lee <[email protected]> wrote regarding Re:
[Radiance-general] The sub-4K second benchmark barrier...:
see:
http://ww.blastwave.org/pacakges/radiance
Sooryy, I Musy leran to tpye:
http://www.blastwave.org/packages/radiance
Checked by use this time. It is nearly the end of my day!
I ran Mark's Benchmark test on my iMac 10.4.7 Core2Dou 2.14 GHz 1 GB RAM
with Radiance 3.8 Universal binary and I got times around 9000 seconds.
Seems too long.
I was hoping it would be in the 4K region. Of course I closed the shell
before copying the times, so I don't have the exact number.
For a Core2Duo, does Radiance use both processors, or would I have to use
rad with -N 2?
I know you have to do that with a render farm or any multiple processor
machine, but since the Core2Duo is 2 on one chip, I'm not sure.
For the benchmark itself, I saw Mark only wants single processor results.
Rob F
···
-----Original Message-----
From: Rob Guglielmetti [mailto:[email protected]]
Sent: Wednesday, November 01, 2006 9:40 AM
To: Radiance general discussion
Cc: Ferdinand Schmid
Subject: [Radiance-general] The sub-4K second benchmark barrier...
...has been broken!
That's right, people. One of the beloved render farm machines at my office
just turned in a sub-4K second timing on the Mark Stock Radiance benchmark
scene. SSSssssssss!
What's interesting to note is that there are subtle differences in cpu
speed, even among the same cpu model. I assume this is due to different
voltage settings on the motherboards, but I'm not positive. I also wonder
how much of the speedup is attributable to the use of the new official
release of Radiance 3.8. Another interesting thing is that three of four
benchmarks had the same exact number of rays traced. I find this a little
odd.
I was hoping Greg and maybe Ferdinand (builder and maintainer of my office's
render farm systems) could weigh in on this. I have a couple other
benchmark timings and stats to add to the site
(http://mark.technolope.org/pages/rad_bench.html) as well, I was hoping
others have installed the official 3.8 and were doing similar testing.
Happy benchmarking.
- Rob
_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
Rob,
As it stands, rpict will only open one process, which will only load up one core.
I would also like to begin collecting multiprocessor benchmarks, but there was once confusion as to how to get a consistent measurement across different platforms. Is rad with "-N" a better or more consistent tool for preparing multiprocessor benchmarks, or is the runsmp script in bench4 better?
I already have a handful of benchmark runs using the runsmp script (though it allows more variables such as NCPU and number of subdivisions, both of which strongly affect runtime).
Or is it likely that rpict itself will soon support pthreads?
Any input would be appreciated.
Mark
[email protected]
···
On Wed, 15 Nov 2006, Fitzsimmons, Rob wrote:
I ran Mark's Benchmark test on my iMac 10.4.7 Core2Dou 2.14 GHz 1 GB RAM
with Radiance 3.8 Universal binary and I got times around 9000 seconds.
Seems too long.
I was hoping it would be in the 4K region. Of course I closed the shell
before copying the times, so I don't have the exact number.
For a Core2Duo, does Radiance use both processors, or would I have to use
rad with -N 2?
I know you have to do that with a render farm or any multiple processor
machine, but since the Core2Duo is 2 on one chip, I'm not sure.
For the benchmark itself, I saw Mark only wants single processor results.
Rob F
-----Original Message-----
From: Rob Guglielmetti [mailto:[email protected]]
Sent: Wednesday, November 01, 2006 9:40 AM
To: Radiance general discussion
Cc: Ferdinand Schmid
Subject: [Radiance-general] The sub-4K second benchmark barrier...
...has been broken!
That's right, people. One of the beloved render farm machines at my office
just turned in a sub-4K second timing on the Mark Stock Radiance benchmark
scene. SSSssssssss!
What's interesting to note is that there are subtle differences in cpu
speed, even among the same cpu model. I assume this is due to different
voltage settings on the motherboards, but I'm not positive. I also wonder
how much of the speedup is attributable to the use of the new official
release of Radiance 3.8. Another interesting thing is that three of four
benchmarks had the same exact number of rays traced. I find this a little
odd.
I was hoping Greg and maybe Ferdinand (builder and maintainer of my office's
render farm systems) could weigh in on this. I have a couple other
benchmark timings and stats to add to the site
(http://mark.technolope.org/pages/rad_bench.html\) as well, I was hoping
others have installed the official 3.8 and were doing similar testing.
Happy benchmarking.
- Rob
_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
Mark is right -- two cores on one chip doesn't mean they both get used automatically. The OS isn't quite that helpful... The runsmp script that uses rpiece (I think?) is better than using rad -N 2, since the latter requires multiple views to even run in parallel.
-Greg
···
From: Mark Stock <[email protected]>
Date: November 15, 2006 1:59:07 PM PST
Rob,
As it stands, rpict will only open one process, which will only load up one core.
I would also like to begin collecting multiprocessor benchmarks, but there was once confusion as to how to get a consistent measurement across different platforms. Is rad with "-N" a better or more consistent tool for preparing multiprocessor benchmarks, or is the runsmp script in bench4 better?
I already have a handful of benchmark runs using the runsmp script (though it allows more variables such as NCPU and number of subdivisions, both of which strongly affect runtime).
Or is it likely that rpict itself will soon support pthreads?
Any input would be appreciated.
Mark
[email protected]
Another Benchmark:
rpict time: 3486.07
processor: Intel Xeon Dual Core (MacPro)
total proc/core: 2/4
clock speed: 2.66 GHz
cache: 4MB/proc
OS/kernel: OSX 10.4.8
RAD version: 3.8
compiler: using precompilied MacOS binary from radiance website
options: ?
results: 694,474,996 rays
date: 11/22/06
submitter: kirk t.
Also, using N=4CPU,
rpict time = 876.30
Allright, added.
It's nice how the number of rays traced doesn't change from -O2 precompiled binaries to the -O3 -fast that Matt used. The precompiled binaries seem quite fast already on Macs.
Those new macs seem quite nice.
Thanks for submitting!
Mark
···
On Wed, 22 Nov 2006, Kirk Thibault wrote:
Another Benchmark:
rpict time: 3486.07
processor: Intel Xeon Dual Core (MacPro)
total proc/core: 2/4
clock speed: 2.66 GHz
cache: 4MB/proc
OS/kernel: OSX 10.4.8
RAD version: 3.8
compiler: using precompilied MacOS binary from radiance website
options: ?
results: 694,474,996 rays
date: 11/22/06
submitter: kirk t.
Also, using N=4CPU,
rpict time = 876.30
_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general