I did the same Radiance simulation, one without specifying the "-n" option
for rtrace, the other with "-n 8" specified for rtrace. The former took 9
min to finish, whereas the latter used 13 min ...
I was quite happy to see all the 8 cores of the CPU are fully utilized in
the beginning, but I was quite upset when the results turned out to be
worse than not specifying anything ...
Why the -n option didn't reduce the simulation time but increases it ? ...
Did I miss something obvious here regarding Radiance parallel rendering ?
The two bash scripts are something like:
oconv ./Sky_m3_d21_h13.rad ./materials.mat ./Context.rad >
./Octree_files_m3_d21_h13.oct
cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar 256
-as 512 ./Octree_files_.oct > ./Results_m3_d21_h13_ir_RGB.txt
and:
oconv ./Sky_m3_d21_h13.rad ./materials.mat ./Context.rad >
./Octree_files_m3_d21_h13.oct
cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar 256
-as 512 -n 8 ./Octree_files_.oct > ./Results_m3_d21_h13_ir_RGB.txt
The test was done on a Macbook Pro, OS X 10.8.2, 2.3 GHz Intel Core i7 (8
cores), with 8GB REM.
The images showing the CPU core usage for the two simulations are put at:
http://episode-hopezh.blogspot.sg/2012/10/rtrace-parallel-rendering-myth.html
Advices are appreciated! and thanks in advance!
- Ji
Hi Ji,
This has been discussed before in an earlier thread -- some months ago, I guess. You don't get a speed-up from parallel processing if you fail to set the -af option to share indirect values. This is because you force each process to recompute the same values redundantly. Add a -af option and see what happens.
Cheers,
-Greg
···
From: Ji Zhang <[email protected]>
Date: October 19, 2012 8:02:47 AM PDT
I did the same Radiance simulation, one without specifying the "-n" option for rtrace, the other with "-n 8" specified for rtrace. The former took 9 min to finish, whereas the latter used 13 min ...
I was quite happy to see all the 8 cores of the CPU are fully utilized in the beginning, but I was quite upset when the results turned out to be worse than not specifying anything ...
Why the -n option didn't reduce the simulation time but increases it ? ...
Did I miss something obvious here regarding Radiance parallel rendering ?
The two bash scripts are something like:
oconv ./Sky_m3_d21_h13.rad ./materials.mat ./Context.rad > ./Octree_files_m3_d21_h13.oct
cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar 256 -as 512 ./Octree_files_.oct > ./Results_m3_d21_h13_ir_RGB.txt
and:
oconv ./Sky_m3_d21_h13.rad ./materials.mat ./Context.rad > ./Octree_files_m3_d21_h13.oct
cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar 256 -as 512 -n 8 ./Octree_files_.oct > ./Results_m3_d21_h13_ir_RGB.txt
The test was done on a Macbook Pro, OS X 10.8.2, 2.3 GHz Intel Core i7 (8 cores), with 8GB REM.
The images showing the CPU core usage for the two simulations are put at:
http://episode-hopezh.blogspot.sg/2012/10/rtrace-parallel-rendering-myth.html
Advices are appreciated! and thanks in advance!
- Ji
Thank you very much, Greg!
Following your advice, the simulation time is reduced from 9 min to just 3
min !
The bash script is:
cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar 256
-as 512 -af ./ambient_file.amb -n 8 ./Octree_files_.oct > ./Results.txt
(I'm very eager to test it on our HPC later ! )
I think this can be one of the FAQs of Radiance.
Thank you again! And have a nice weekend!
-Ji
···
On Friday, October 19, 2012, Greg Ward wrote:
Hi Ji,
This has been discussed before in an earlier thread -- some months ago, I
guess. You don't get a speed-up from parallel processing if you fail to
set the -af option to share indirect values. This is because you force
each process to recompute the same values redundantly. Add a -af option
and see what happens.
Cheers,
-Greg
> From: Ji Zhang <[email protected] <javascript:;>>
> Date: October 19, 2012 8:02:47 AM PDT
>
> I did the same Radiance simulation, one without specifying the "-n"
option for rtrace, the other with "-n 8" specified for rtrace. The former
took 9 min to finish, whereas the latter used 13 min ...
>
>
>
> I was quite happy to see all the 8 cores of the CPU are fully utilized
in the beginning, but I was quite upset when the results turned out to be
worse than not specifying anything ...
>
>
>
> Why the -n option didn't reduce the simulation time but increases it ?
...
>
>
>
> Did I miss something obvious here regarding Radiance parallel rendering ?
>
>
> The two bash scripts are something like:
>
> oconv ./Sky_m3_d21_h13.rad ./materials.mat ./Context.rad >
./Octree_files_m3_d21_h13.oct
>
> cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar
256 -as 512 ./Octree_files_.oct > ./Results_m3_d21_h13_ir_RGB.txt
>
>
>
> and:
>
>
>
> oconv ./Sky_m3_d21_h13.rad ./materials.mat ./Context.rad >
./Octree_files_m3_d21_h13.oct
>
> cat ./sensor.txt | rtrace -I -h -w -oov -u- -aa 0.1 -ab 6 -ad 1024 -ar
256 -as 512 -n 8 ./Octree_files_.oct > ./Results_m3_d21_h13_ir_RGB.txt
>
>
>
> The test was done on a Macbook Pro, OS X 10.8.2, 2.3 GHz Intel Core i7
(8 cores), with 8GB REM.
>
>
>
> The images showing the CPU core usage for the two simulations are put at:
>
>
http://episode-hopezh.blogspot.sg/2012/10/rtrace-parallel-rendering-myth.html
>
>
>
> Advices are appreciated! and thanks in advance!
> - Ji
_______________________________________________
Radiance-general mailing list
[email protected] <javascript:;>
http://www.radiance-online.org/mailman/listinfo/radiance-general