FLuid simulation question

The interface, modeling, 3d editing tools, import/export, feature requests, etc

Moderators: jesterKing, stiv

Post Reply
Amok
Posts: 0
Joined: Mon Mar 03, 2008 11:21 pm
Location: Romania Iasi

FLuid simulation question

Post by Amok »

Hi. I'm new here, even if not new to 3d graphics. I have 2 machines, one with 2 Xeon quads and 1 with 1 P4 at 3.2 Ghz, which i have doing nothing in my house.. My main machine is fast, it's amazing at calculating simulations in a manageable time frame, but i was wondering if it actually uses all the cores when making the simulation?

Since i would love the idea of letting the old one make the sim, while i go on with my tasks on the main machine. So, to be more specific

Does the fluid simulation take advantage of all the 8 cores? Because i am using the low machine for night time simulations, but i can't decide if i should do the same during the day, while i work at my projects.

Thank you in advance.

jesterKing
Site Admin
Posts: 207
Joined: Fri Oct 18, 2002 12:48 pm
Location: Finland

Post by jesterKing »

You need an OpenMP-enabled build. If you build with scons, you can enable it with WITH_BF_OPENMP=1 on the command-line:

scons WITH_BF_OPENMP=1

or set it in your user-config.py:

WITH_BF_OPENMP='true'

If you're building with msvc, then you need a professional license. If you're compiling with GCC, you need at least 4.2

/Nathan

Amok
Posts: 0
Joined: Mon Mar 03, 2008 11:21 pm
Location: Romania Iasi

Post by Amok »

jesterKing wrote:You need an OpenMP-enabled build. If you build with scons, you can enable it with WITH_BF_OPENMP=1 on the command-line:

scons WITH_BF_OPENMP=1

or set it in your user-config.py:

WITH_BF_OPENMP='true'

If you're building with msvc, then you need a professional license. If you're compiling with GCC, you need at least 4.2

/Nathan
I have very low experience as a programer, i blame Bill for that :). So i would probably fail at applying your advice, though i thank you for it.

My question was: Is the standard fluid simulation in Blender 2.45 uses all the cores to calculate or just 1? or how many. Since i wonder if it's more efficient to wait for my main machine to calculate or to get the scene on the secondary machine and calculate it there, while i do my normal tasks on the main.

jesterKing
Site Admin
Posts: 207
Joined: Fri Oct 18, 2002 12:48 pm
Location: Finland

Post by jesterKing »

What I implicitely tried to say was that Blender 2.45 doesn't have OpenMP support built in, so it won't utilise all available cores when running a fluid simulation. You can get it with current SVN sources, so that's why I said that you could build it :) I *think* I have an openmp build available here: http://letworyinteractive.com/blender/c ... openmp.exe , but I'm not sure, because with one openmp build I forgot to actually turn support on. Anyway, try it and see if all cores get utilised. If they do, then it is most efficient to run it on the machine with most cores :)

/Nathan

Amok
Posts: 0
Joined: Mon Mar 03, 2008 11:21 pm
Location: Romania Iasi

Post by Amok »

jesterKing wrote:What I implicitely tried to say was that Blender 2.45 doesn't have OpenMP support built in, so it won't utilise all available cores when running a fluid simulation. You can get it with current SVN sources, so that's why I said that you could build it :) I *think* I have an openmp build available here: http://letworyinteractive.com/blender/c ... openmp.exe , but I'm not sure, because with one openmp build I forgot to actually turn support on. Anyway, try it and see if all cores get utilised. If they do, then it is most efficient to run it on the machine with most cores :)

/Nathan
thanks a lot. since i have 8 cores, i'll test this tomorow with both standard 2.45 and your build, on a medium scene and i'll make a topic, to give people a kind of a benchmark to see how multi cores can affect the calculation.

FuzzCube
Posts: 0
Joined: Thu Mar 20, 2008 5:56 am
Location: at my desk

Post by FuzzCube »

ok since this is a simmilar subject i thought i should add it to this topic....

I am building a small render farm of the same (dual 771 xeon quad cores) x4 machines minimum running 64bit linux...

will the support for such as this be implemented in blender 2.50? or earlier versions


and also does blender support nvidia Quadro FX mid end cards 1500-1700?

jesterKing
Site Admin
Posts: 207
Joined: Fri Oct 18, 2002 12:48 pm
Location: Finland

Post by jesterKing »

FuzzCube wrote:ok since this is a simmilar subject i thought i should add it to this topic....

I am building a small render farm of the same (dual 771 xeon quad cores) x4 machines minimum running 64bit linux...

will the support for such as this be implemented in blender 2.50? or earlier versions


and also does blender support nvidia Quadro FX mid end cards 1500-1700?
If you can get builds with OpenMP enabled for 64bit linux, you should be fine.

Blender runs on any card that supports OpenGL 1.1 (barring driver issues). The GPU is not used for offloading rendering, so don't expect better render times through upgrading video cards. At most, the interface might be even more responsive ;)

/Nathan

FuzzCube
Posts: 0
Joined: Thu Mar 20, 2008 5:56 am
Location: at my desk

Post by FuzzCube »

Thanks, JesterKing, :D

you've helped me helped me with a big decision here as a quadro 1500FX costs 500$ + versus a 570FX which costs around 200$

Since the render engine wont take advantage of this then I don't see any real need to load up on more expensive cards for my workstations, so this will save me some cash and I can get them built faster as each machine wont cost nearly as much so thanks again!!! :D

(also as a side note it seems that nvidia put more descriptive information on their website as comparison to quadroFX versus consumer graphics, which also helps too because this in general is what I am looking for as the need for better quality mesh preview (before rendering in edit and object mode) is often needed in my projects, as renders can take a while even with multi cores and it's annoying to find that a setting wasn't right after waiting for a render to finish and then having to re render to fix one tiny setting or whatever.

jesterKing
Site Admin
Posts: 207
Joined: Fri Oct 18, 2002 12:48 pm
Location: Finland

Post by jesterKing »

FuzzCube wrote:Thanks, JesterKing, :D

you've helped me helped me with a big decision here as a quadro 1500FX costs 500$ + versus a 570FX which costs around 200$

Since the render engine wont take advantage of this then I don't see any real need to load up on more expensive cards for my workstations, so this will save me some cash and I can get them built faster as each machine wont cost nearly as much so thanks again!!! :D

(also as a side note it seems that nvidia put more descriptive information on their website as comparison to quadroFX versus consumer graphics, which also helps too because this in general is what I am looking for as the need for better quality mesh preview (before rendering in edit and object mode) is often needed in my projects, as renders can take a while even with multi cores and it's annoying to find that a setting wasn't right after waiting for a render to finish and then having to re render to fix one tiny setting or whatever.
I'm glad I could be of service :)

/Nathan

Post Reply