 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
with compiler options (and possibly installing a new OS)! All on an AMD Phenom
II x6 1090T @ stock (3.2 GHz). Still plan on installing FreeBSD on this computer
eventually just for Povray speed improvements...
FreeBSD 8.2, gcc 4.6, -march=barcelona
Render Time:
Photon Time: 0 hours 0 minutes 2 seconds (2.390 seconds)
using 9 thread(s) with 2.763 CPU-seconds total
Radiosity Time: No radiosity
Trace Time: 0 hours 3 minutes 10 seconds (190.466 seconds)
using 6 thread(s) with 1113.568 CPU-seconds total
Debian 7.0, gcc 4.6.1, -march=barcelona -mfpmath=both
Render Time:
Photon Time: 0 hours 0 minutes 2 seconds (2.300 seconds)
using 9 thread(s) with 2.663 CPU-seconds total
Radiosity Time: No radiosity
Trace Time: 0 hours 3 minutes 17 seconds (197.996 seconds)
using 6 thread(s) with 1181.504 CPU-seconds total
Debian 7.0, gcc 4.6.1, -march=barcelona
Render Time:
Photon Time: 0 hours 0 minutes 2 seconds (2.277 seconds)
using 9 thread(s) with 2.648 CPU-seconds total
Radiosity Time: No radiosity
Trace Time: 0 hours 3 minutes 38 seconds (218.326 seconds)
using 6 thread(s) with 1277.363 CPU-seconds total
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
jhu <nomail@nomail> wrote:
> FreeBSD 8.2, gcc 4.6, -march=barcelona
You can use "-march=native" to tell gcc to automatically optimize for
the current system.
Btw, what other optimization options are you using? I assume that at
least "-O3"? Besides that, try using "-ffast-math". You might also try
"-mfpmath=sse -msse2" (or a higher number if gcc supports it).
--
- Warp
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Warp <war### [at] tag povrayorg> wrote:
> jhu <nomail@nomail> wrote:
> > FreeBSD 8.2, gcc 4.6, -march=barcelona
>
> You can use "-march=native" to tell gcc to automatically optimize for
> the current system.
>
> Btw, what other optimization options are you using? I assume that at
> least "-O3"? Besides that, try using "-ffast-math". You might also try
> "-mfpmath=sse -msse2" (or a higher number if gcc supports it).
>
> --
> - Warp
Forgot to mention that this is on 64-bit Debian and FreeBSD, so -msse2 is
implicitly set. Other options are the usual ones that configure provides (that
includes -O3 and -ffast-math). What I find interesting is that Povray on FreeBSD
is faster without the "-mfpmath=both" (times not listed) whereas on Linux
"-mfpmath=both" significantly speeds up rendering to within 3% of the FreeBSD
time. Must be something with the kernels of each OS.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
jhu wrote:
> with compiler options (and possibly installing a new OS)! All on an AMD Phenom
> II x6 1090T @ stock (3.2 GHz). Still plan on installing FreeBSD on this computer
> eventually just for Povray speed improvements...
Interesting. Still, I hope you enjoy playing with OS installs and
compiler options as a fun activity in itself, because you probably have
to render a lot of images to regain the time spent on twiddling ;)
Also I wonder how the times would compare to prebuilt windows
binaries on the same machine.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Christian Froeschlin <chr### [at] chrfrde> wrote:
> jhu wrote:
>
> > with compiler options (and possibly installing a new OS)! All on an AMD Phenom
> > II x6 1090T @ stock (3.2 GHz). Still plan on installing FreeBSD on this computer
> > eventually just for Povray speed improvements...
>
> Interesting. Still, I hope you enjoy playing with OS installs and
> compiler options as a fun activity in itself, because you probably have
> to render a lot of images to regain the time spent on twiddling ;)
>
> Also I wonder how the times would compare to prebuilt windows
> binaries on the same machine.
I'm somewhat in the middle of installing various OSes and trying to build povray
3.7 with them. I have FreeBSD figured out; now to see if the other *BSDs are
similar in this respect. Come to think of it, I should just use virtual machines
for these things.
Also I probably would have wasted the time surfing the internet anyway. Damn you
ADHD!!!
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
jhu wrote:
> with compiler options (and possibly installing a new OS)! All on an
> AMD Phenom II x6 1090T @ stock (3.2 GHz). Still plan on installing
> FreeBSD on this computer eventually just for Povray speed
> improvements...
Did you read up on the schedulers on these various operating systems?
They can often be configured to your needs in case the OS doesn't choose
the best scheduling parameters itself. On Linux, you find documentation
about the scheduler in the documentation of your kernel. The tool
"schedtool" is useful and in its manpage, it is noted:
"SCHED_BATCH [ since 2.6.16 in mainline ] SCHED_BATCH was designed for
non-interactive, CPU-bound applications. It uses longer timeslices (to
better exploit the cache), but can be interrupted anytime by other
processes in other classes to guaratee interac‐tion of the system.
Processes in this class are selected last but may result in a
considerable speed-up (up to 300%). No interactive boosting is done."
(On Debian, you will find schedtool in the packages with the same name.
As far as I know you can also set the scheduler parameters via /proc)
So if povray is already running you could do something like
schedtool -3 `pidof povray`
or to start povray using SCHED_BATCH:
schedtool -3 -e povray
In theory, this should speed up things if your running other processes
besides POVRay.
Greetings
Urs
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Urs Holzer <urs### [at] andonyarcom> wrote:
> jhu wrote:
> > with compiler options (and possibly installing a new OS)! All on an
> > AMD Phenom II x6 1090T @ stock (3.2 GHz). Still plan on installing
> > FreeBSD on this computer eventually just for Povray speed
> > improvements...
>
> Did you read up on the schedulers on these various operating systems?
> They can often be configured to your needs in case the OS doesn't choose
> the best scheduling parameters itself. On Linux, you find documentation
> about the scheduler in the documentation of your kernel. The tool
> "schedtool" is useful and in its manpage, it is noted:
>
> "SCHED_BATCH [ since 2.6.16 in mainline ] SCHED_BATCH was designed for
> non-interactive, CPU-bound applications. It uses longer timeslices (to
> better exploit the cache), but can be interrupted anytime by other
> processes in other classes to guaratee interac‐tion of the system.
> Processes in this class are selected last but may result in a
> considerable speed-up (up to 300%). No interactive boosting is done."
>
> (On Debian, you will find schedtool in the packages with the same name.
> As far as I know you can also set the scheduler parameters via /proc)
>
> So if povray is already running you could do something like
> schedtool -3 `pidof povray`
> or to start povray using SCHED_BATCH:
> schedtool -3 -e povray
>
> In theory, this should speed up things if your running other processes
> besides POVRay.
>
> Greetings
> Urs
I wasn't aware of that, although my computer is running headless. All other
processes are at idle.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
