Page 1 of 2

Processing Geometry 1-Core Hold up

Posted: Fri Nov 16, 2012 6:29 pm
by leblanc980
I'm not sure if this problem is on 3ds max side, or octanes, but when I click the render button (The regular way, not through octanes viewport) the first process performed is called "Processing Geometry". I guess it is taking all of the geometry from the scene, and loading it to octane's engine to be rendered.

My problem is it looks like this first process only uses 1 core. For a scene I'm working on that is 2.8 millon verts this takes over a minute for the "Processing Geometry" to occur. Now that I am animating, each new frame does it again and again... Taking forever.
If this could use all four cores instead of one, it would make more sense.

I'm using max 2012, on Octane RC2.

Any Ideas???

Thanks in advance!

Re: Processing Geometry 1-Core Hold up

Posted: Fri Nov 16, 2012 6:51 pm
by face
Correct, Octane uses only one core to send data to the GPU.
There isn“t any other way yet...

face

Re: Processing Geometry 1-Core Hold up

Posted: Fri Nov 16, 2012 7:03 pm
by leblanc980
Bummer.

I gotta say I was scared that my CPU blew up there for a minute LOL.. Not a huge deal, but I hope these guys can figure that out in the future.

Thanks!

Re: Processing Geometry 1-Core Hold up

Posted: Sun Nov 18, 2012 12:57 am
by leblanc980
I found out alot of programs use only one core, and did some research as a means to try and figure out a way to speed up this process. I came across disabling hyper threading which pretty much would turned my 4 core processor into a 2 core processor. It took me a while to figure it all out in my bios as it wasn't called "Hyper Threading", but rather called core1 or something on my Biostar TA990FXE.

The theory behind it is....
I'm using 1 of my 4 cores at 100% which equals 25% of my cpu,
So than using 1 of 2 cores at 100% would be 50% of my cpu and there for cut in half the time it takes 3ds Max to process.

I ran tests on the same scene and found out that using this method which disables hyper threading turning my 4-core into 2-cores actually took longer to process the geometry and render the scene. So it literally back fired in my face!

I'm clueless to why this is the way it is. Perhaps my overclocked DDR memory acted differently at the 2-core level??? Causing a bottle neck of some sort??

Back to the original 4-cores for now. Thought I would share.

Re: Processing Geometry 1-Core Hold up

Posted: Tue Nov 20, 2012 9:00 am
by suvakas
leblanc980 wrote:I found out alot of programs use only one core, and did some research as a means to try and figure out a way to speed up this process. I came across disabling hyper threading which pretty much would turned my 4 core processor into a 2 core processor. It took me a while to figure it all out in my bios as it wasn't called "Hyper Threading", but rather called core1 or something on my Biostar TA990FXE.

The theory behind it is....
I'm using 1 of my 4 cores at 100% which equals 25% of my cpu,
So than using 1 of 2 cores at 100% would be 50% of my cpu and there for cut in half the time it takes 3ds Max to process.

I ran tests on the same scene and found out that using this method which disables hyper threading turning my 4-core into 2-cores actually took longer to process the geometry and render the scene. So it literally back fired in my face!

I'm clueless to why this is the way it is. Perhaps my overclocked DDR memory acted differently at the 2-core level??? Causing a bottle neck of some sort??

Back to the original 4-cores for now. Thought I would share.
Are you serious?
How can this make your CPU to go any faster?

Suv

Re: Processing Geometry 1-Core Hold up

Posted: Thu Nov 22, 2012 12:54 am
by leblanc980
Simple Math.

Re: Processing Geometry 1-Core Hold up

Posted: Thu Nov 22, 2012 2:51 am
by TBFX
leblanc980 wrote:Simple Math.
Yes, and it's strange that it didn't help you. I have a 4 physical core i7 that was originally hyperthreaded to 8 logical cores, so mesh read and voxelization (in standalone) was only using 12.5% of my CPU. I turned off hyperthreading so a single threaded task would use one whole physical core ie. 25% of CPU and my load/voxelization times were halved!

I don't use MAX and haven't even though to try turning hyperthreading back on running the Maya plugin as like you say simple math says what I have will be faster.

EDIT: Had a thought. Transfer/Voxelization can eat a lot of system RAM. Maybe when you utilized more CPU it increased your RAM usage to the point of swapping and that's why the performance dropped?

T.

Re: Processing Geometry 1-Core Hold up

Posted: Thu Nov 22, 2012 4:43 am
by leblanc980
@TBFX
I'm totally blown away I didn't see any difference. At first I thought I literally turned off two of my cores, but my CPU-Z said I was still running 4.5Ghz, so that couldn't have been the case. I agree and think it has something to do with my memory. At this point, I will probably screw up my bios even more than its worth to try and figure out a solution lol. My Biostar TA 990FXE motherboard bios is kind of tricky

Re: Processing Geometry 1-Core Hold up

Posted: Thu Nov 22, 2012 5:29 am
by roeland
TBFX wrote: I turned off hyperthreading so a single threaded task would use one whole physical core ie. 25% of CPU and my load/voxelization times were halved!
Did you measure this? While windows sees 8 cores when hyperthreading is enabled, there are still only 4 physical cores, and one thread still uses 100% of a physical core. I would expect the voxelization to be approximately equally fast with or without hyperthreading. The display of Windows task manager assumes 8 independent cores which can be misleading.

--
Roeland

Re: Processing Geometry 1-Core Hold up

Posted: Thu Nov 22, 2012 5:37 am
by TBFX
roeland wrote:Did you measure this? While windows sees 8 cores when hyperthreading is enabled, there are still only 4 physical cores, and one thread still uses 100% of a physical core. I would expect the voxelization to be approximately equally fast with or without hyperthreading. The display of Windows task manager assumes 8 independent cores which can be misleading.
Absolutely, It was a year ago when I did this using the standalone but CPU usage went from 12.5% with hyperthreading enabled and jumped up to 25% once I turned it off. Load/voxelization time on my scene (which almost filled the 3GB Vram) went from 16 min down to between 8 and 9 min.

EDIT: I've seen this behavior with other single threaded software or tasks which is why I gave it a try. From all evidence I've seen when hyperthreading is enabled a single threaded process only uses the equivalent to one logical core, this can of course be spread across several cores.

T.