I know there are memory limitations that restrict overall scene size.
1. Does octane intelligently cull polys that aren't visible to camera, or are they still loaded to the GPU?
2. Do multiple cards (and hence VRAM) increase the poly limit, or is the poly count limited to what can fit on any one card in the system?
3. What happens when you hit the poly limit? Does octane fall over, or does it just tell you things got to big...?
Thanks.
Scene size limitations with multiple GPU cards
Forum rules
For new users: this forum is moderated. Your first post will appear only after it has been reviewed by a moderator, so it will not show up immediately.
This is necessary to avoid this forum being flooded by spam.
For new users: this forum is moderated. Your first post will appear only after it has been reviewed by a moderator, so it will not show up immediately.
This is necessary to avoid this forum being flooded by spam.
1. All polys are loaded. With Octane you will find you want to move the camera around the scene in 'realtime' and make a bracket of interesting shots so its good to have a full mesh anyway.
2. For multiple cards the memory available for Octane is the least common amount as the scene is loaded in both ie. if you have 2 x 1gb cards you have 1gb to use. If you use a 1gb and 1.5gb card you have 1gb to use. Having 2 identical cards is best else the faster one waits some on the slower one. You need to use Nvidia Cuda capable cards only. ATI and Nvidia together doesnt work. ATI and OpenCL is not supported yet and probably wont be for some time.
Practical performance is improved by using a low cost card dedicated to UI display and another grunt card(s) without monitors connected dedicated to Cuda rendering.
3. If you try to load a scene that overfills the available vram you should get a popup warning of that at the time. You will then need to abandon loading and attend to your scene to reduce the space required. Often you can make more intelligent use of textures to get around a shortfall.
In practice about 3m polys per gb is an estimate that allows for textures and your image - it depends on how you trade off the space. If you have a lot of textures or a large image resolution obviously you have less room for polys.
If you have a need to do very complex scenes at very high res (like for print or posters) then you will need to look at multiple (for speed) Quadro or Tesla cards with 3,4 or 6 gb vram each.
There are some gaming cards with 1.5, 1.8 or 2 gb vram which is a handy size and these are quite suited to many Octane tasks and come at a much better price.
Quadros with 2 and 2.5gb vram give a bit extra space but these are relatively expensive and not as fast as top end gaming cards.
A gaming card and a Quadro with the same number of Cuda cores perform much the same in Octane but the Quadro typically has more space.
For less demanding use 896mb-1.28gb cards are ok. Many users have these and they are fine for their purposes. Cards with 96, 112 or 128 Cuda cores are about the minimum you would want to bother with.
The new Fermi 460/1gb/2gb, 465, 470, and 480 are popular choices.
The GTX460/2gb presently is a good combination of vram capacity, decent performance, moderate cost, reasonable power draw, and is quiet and cool compared to a 470 or 480. A 460 gives about half the performance of a 480.
If you want to use multiple cards make sure you have a quality power supply with enough connectors and enough Watts to run them. This means about 700W-1200W or more. Your motherboard needs enough pci-e slots and appropriately spaced to fit the video cards. pci-e16x and 8x are best, 4x will do, 1x isnt suitable. An alternative solution is to use an expander box with its own power supply. See the GPU-X link in the header.
SLI isnt used by Octane and needs to be disabled. In general dont overclock your cards as it can cause instability and overheating with full on Octane use.
To fully load a card that has a large vram (896mb+) with polys you really need 64bit and 4gb+ system memory and possibly 6,8 or 12gb depending on your card(s) capacity.
32 bit is limited to about 1.7 gb system memory use in practice (or 2.7gb with the 3gb switch if you have XP pro) however its ok for use with cards with lesser vram like 512, 768 or 896mb.
These sizes are a little restrictive for Octane purposes though.
32 bit will still load Octane into the bigger cards but it wont be able to push in as many polys as with 64 bit.
HTH
2. For multiple cards the memory available for Octane is the least common amount as the scene is loaded in both ie. if you have 2 x 1gb cards you have 1gb to use. If you use a 1gb and 1.5gb card you have 1gb to use. Having 2 identical cards is best else the faster one waits some on the slower one. You need to use Nvidia Cuda capable cards only. ATI and Nvidia together doesnt work. ATI and OpenCL is not supported yet and probably wont be for some time.
Practical performance is improved by using a low cost card dedicated to UI display and another grunt card(s) without monitors connected dedicated to Cuda rendering.
3. If you try to load a scene that overfills the available vram you should get a popup warning of that at the time. You will then need to abandon loading and attend to your scene to reduce the space required. Often you can make more intelligent use of textures to get around a shortfall.
In practice about 3m polys per gb is an estimate that allows for textures and your image - it depends on how you trade off the space. If you have a lot of textures or a large image resolution obviously you have less room for polys.
If you have a need to do very complex scenes at very high res (like for print or posters) then you will need to look at multiple (for speed) Quadro or Tesla cards with 3,4 or 6 gb vram each.
There are some gaming cards with 1.5, 1.8 or 2 gb vram which is a handy size and these are quite suited to many Octane tasks and come at a much better price.
Quadros with 2 and 2.5gb vram give a bit extra space but these are relatively expensive and not as fast as top end gaming cards.
A gaming card and a Quadro with the same number of Cuda cores perform much the same in Octane but the Quadro typically has more space.
For less demanding use 896mb-1.28gb cards are ok. Many users have these and they are fine for their purposes. Cards with 96, 112 or 128 Cuda cores are about the minimum you would want to bother with.
The new Fermi 460/1gb/2gb, 465, 470, and 480 are popular choices.
The GTX460/2gb presently is a good combination of vram capacity, decent performance, moderate cost, reasonable power draw, and is quiet and cool compared to a 470 or 480. A 460 gives about half the performance of a 480.
If you want to use multiple cards make sure you have a quality power supply with enough connectors and enough Watts to run them. This means about 700W-1200W or more. Your motherboard needs enough pci-e slots and appropriately spaced to fit the video cards. pci-e16x and 8x are best, 4x will do, 1x isnt suitable. An alternative solution is to use an expander box with its own power supply. See the GPU-X link in the header.
SLI isnt used by Octane and needs to be disabled. In general dont overclock your cards as it can cause instability and overheating with full on Octane use.
To fully load a card that has a large vram (896mb+) with polys you really need 64bit and 4gb+ system memory and possibly 6,8 or 12gb depending on your card(s) capacity.
32 bit is limited to about 1.7 gb system memory use in practice (or 2.7gb with the 3gb switch if you have XP pro) however its ok for use with cards with lesser vram like 512, 768 or 896mb.
These sizes are a little restrictive for Octane purposes though.
32 bit will still load Octane into the bigger cards but it wont be able to push in as many polys as with 64 bit.
HTH
Last edited by pixelrush on Sun Aug 22, 2010 4:31 pm, edited 20 times in total.
i7-3820 @4.3Ghz | 24gb | Win7pro-64
GTS 250 display + 2 x GTX 780 cuda| driver 331.65
Octane v1.55
GTS 250 display + 2 x GTX 780 cuda| driver 331.65
Octane v1.55
We try to be helpful here. 
I edited it 'a few' times to give a fuller answer so its covers FAQ people might have.
I edited it 'a few' times to give a fuller answer so its covers FAQ people might have.
i7-3820 @4.3Ghz | 24gb | Win7pro-64
GTS 250 display + 2 x GTX 780 cuda| driver 331.65
Octane v1.55
GTS 250 display + 2 x GTX 780 cuda| driver 331.65
Octane v1.55