 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Let me apologize ahead of time for not being at all specific, but when I
compared 3.62, 32 vs. 64 bit, I saw roughly a factor of two improvement in
elapsed time for the 64 bit. With beta 36, 32 vs. 64, the improvement is only
10%. (I'm running Win7-64 on a i7-860.)
Anyone seeing similar results?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Let me add a little bit to this: the factor of two (or more) is for one
particular scene--the classic floating glass sphere--but I'm only seeing the
improvement with 3.6.2. Another scene shows no such 64-bit improvement. And it's
not photons--the large factor is there with or without. But I'm not seeing
anything like that 64-bit advantage in 3.7: just 10%, or so.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
sevendy a écrit :
> Let me apologize ahead of time for not being at all specific, but when I
> compared 3.62, 32 vs. 64 bit, I saw roughly a factor of two improvement in
> elapsed time for the 64 bit. With beta 36, 32 vs. 64, the improvement is only
> 10%. (I'm running Win7-64 on a i7-860.)
>
> Anyone seeing similar results?
I have no time to test both on windows 32 & 64 bits (more over, my
hardware would be different as well as the OS: 2008R2 for 64, XP for 32)
But could you please report all the time (from the reported renders), so
that we could at least have an idea of your glose ?
For instance, if 3.62 32 was 2 hours, 3.62 64 1 hour, and 3.7b36 32/64
is roughly 1 hour, I do not see any issue.
If 3.7b36 32/64 is now 8 days, now that's a problem!
Please also report CPU, frequency & memory (total & available during
render, including used-swap size, as reported by task manager)
--
A good Manager will take you
through the forest, no mater what.
A Leader will take time to climb on a
Tree and say 'This is the wrong forest'.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Thanks for your response.
FYI:
i7-860 2.8GHz, 4 core, 8 "logical" CPU, 8GB RAM (6.5GB available while
rendering, so RAM to spare). Two cases, both 1024 x 768 AA .3, no radiosity:
time in seconds -- elapsed time (total CPU)
trace Case 1 Case 2
threads CPU(approx.) photons no photons no photons
3.6.2 64bit 13% 99 31 158
3.6.2 32bit, 13% 209 71 160
SSE2
3.7 b36 64bit 1 13% 119(119) 36(36) 154(154)
4 51% 48(159) 13(51) 53(208)
8 100% 37(214) 9(66) 35(270)
3.7 b36 32bit 8 100% 39(228) 10(70) 41(310)
Case 1 is a glass sphere, in front of an image on a plane, illuminated from
above, and casting a shadow on a horizontal plane. Case 2 is a single object
against a distant background.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
sevendy schrieb:
> trace Case 1 Case 2
> threads CPU(approx.) photons no photons no photons
>
> 3.6.2 64bit 13% 99 31 158
>
> 3.6.2 32bit, 13% 209 71 160
> SSE2
>
> 3.7 b36 64bit 1 13% 119(119) 36(36) 154(154)
> 4 51% 48(159) 13(51) 53(208)
> 8 100% 37(214) 9(66) 35(270)
>
> 3.7 b36 32bit 8 100% 39(228) 10(70) 41(310)
To me, these figures are evidence that...
(1) with 3.7, the 64-bit version has a slight speed benefit over the
32-bit version (but not much), which appears to me to make sense,
provided that the 32-bit version uses SSE2;
(2) the 64-bit version of 3.7 is typically slightly less CPU-efficient
(as can be seen when running on a single thread) than that of 3.6, which
makes sense given that 3.7 needs some overhead to make sure that it can
make use of multiple cores (or CPUs);
(3) the 32-bit SSE2 version of 3.6 is exceptionally slow with photons.
From my seat, only (3) opens up questions. My first guess would be that
3.6.2 32-bit SSE2 is compiled differently than 3.7.b36 32-bit - possibly
due to a different compiler, though I'm not sure about that. Would you
mind posting the exact version info (including platform and compiler
identifiers) for each of the POV-Ray versions report when started? You
should find it in the Messages pane, where it will output something like
this:
-------------
Persistence of Vision Raytracer(tm) for Windows.
POV-Ray for Windows is part of the POV-Ray(tm) suite of programs.
This is version 3.7.0.beta.34.msvc8.win64.
[...]
-------------
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
This is version 3.7.0.beta.36.msvc9.win64.
This is version 3.6.2.msvc9.win64.
This is version 3.7.0.beta.36.msvc9-sse2.win32.
This is version 3.6.2.msvc9-sse2.win32.
(I think I'm wearing our the "install" and "uninstall" buttons.)
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Updated with additional 32-bit case 1 w/o photons runs. Note improvement for
single thread vs. 3.6.2:
time in seconds -- elapsed time (total CPU)
trace Case 1 Case 2
threads CPU(approx.) photons no photons no photons
3.6.2 64bit 13% 99 31 158
3.6.2 32bit, 13% 209 71 160
SSE2
3.7 b36 64bit 1 13% 119(119) 36(36) 154(154)
4 51% 48(159) 13(51) 53(208)
8 100% 37(214) 9(66) 35(270)
3.7 b36 32bit 1 37(37)
SSE2 4 13(52)
8 100% 39(228) 10(70) 41(310)
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
sevendy schrieb:
> Updated with additional 32-bit case 1 w/o photons runs. Note improvement for
> single thread vs. 3.6.2:
Again, to me it looks more like 3.6.2 32-bit SSE2 has /some/ issue, as
all the other versions roughly agree on how long Case 1 should take to
render in a single thread. Obviously not compiler-related though.
Is there any special feature you're using in that scene?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Off hand, I just don't know enough about POV-Ray to speculate about where the
problem might be. I'll just stick with the 64-bit versions---why not?
Thanks everyone for your help and expertise.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
