Video card usage distribution

[Moonhowl]

Hello,

first of all I've been eyeing this candy for a while now and I love the price you've put on (and the quality and speed of renders!), even though it's beta, imo it's well worth it.

Now to my question;

I have a single quadro video card and as stated in your FAQ octane takes over it totally so I cant do much else while it renders, you also suggest it might be wise to buy additional video card for "other stuff" and dedicate one to octane.

Fine, but i've been thinking more along the lines of buying another card and dedicate BOTH to octane, but still, I'd like to have just enough resources freed to also do other tasks on my computer while render is running.

Have you been thinking or planning to make it possible to dedicate certain % of resources (if it's even possible, i'm no coder so this is just a shot in the dark:-) for the system to use so it can operate normally while octane renders? like restrict some gpus on the video card for windows solely or smtn?

I think this is certainly an issue to be tackled if it is possible at all, at least it would be most useful for me, since it's most practical to be able to continue modelling in the background while renders run.

[radiance]

Hi,

Currently it's not possible, cuda does'nt allow you to share your GPU and render kernels.
I'd recommend for you to buy a GTX480 or GTX470 as a 2nd headless GPU for use with octane,
and use your current card for display/work...

Radiance

[Moonhowl]

Thanks for the reply!

[havensole]

In my rendering rig I have 2 gtx470's and the motherboard has a built on vga card. Works amazingly fast. When using my system at home which only has the single gts250 everything crawls. I've thought about buying a low end pci based video card just to handle the desktop graphics stuff as my motherboard only has a single pcie slot.

[Moonhowl]

Is it possible to separately use the in-built card while octane is rendering on main card, for example? I'm not sure but I think my MB has one too, I mean most of them do nowadays, I actually havent thought of using it until now^^

[havensole]

I am not sure if I completely understand your question. Are you asking if you can use the built-on card to render with while the other one (main as you put it) is rendering also? Typically the built on cards are pretty worthless to Octane. The one in my system that I mentioned before isn't even Cuda enabled. I have seen many that are, but they have less cores then I are worth trying. The way my system is setup is that the built-on card is my main display card. The other 2 gtx470's are independent of each other (non-sli) and are not hooked up to any monitors. They can render in either a shared configuration, where one scene is rendered using both cards/speed boost, or render 2 different scenes at the same time with two instances of octane open. Having the built on card just makes navigating the OS and Octane a lot smoother. I don't sue this system for anything other then simple stuff and Octane, so I'm not too troubled by the lacking abilities of the built on card. It takes a little to get the OS to like this setup, and some BIOS tweaking, but it is worth it.

[radiance]

we did a test today between a system with 2x Quadro FX3800 in 2x PCI-E slots straight on the motherboard. (Asus WS)
we also did a test on a xpander box with 2x Quadro FX3800 both shares by an 8x PCI-E slot. (eg 2 cards on one 8x bus)

we did'nt notice any difference in speed between both.
it just took a few seconds longer to initially load the model/scene into memory after clicking the meshnode.

Radiance

[DayVids]

we did a test today between a system with 2x Quadro FX3800 in 2x PCI-E slots straight on the motherboard. (Asus WS)
we also did a test on a xpander box with 2x Quadro FX3800 both shares by an 8x PCI-E slot. (eg 2 cards on one 8x bus)

we did'nt notice any difference in speed between both.
it just took a few seconds longer to initially load the model/scene into memory after clicking the meshnode.

Radiance

Correct me if I'm wrong, but that would make perfect sense since the information is loaded into the VRAM, and so there's very little that has to be done outside the card, the only time there would be info transferring to and from, would be when there are initial mesh's being loaded onto the card, and if you had a lot of UV Maps or images that had to be loaded also, it could take longer yet, but once the info is loaded on the card, there's not a lot it has to do, off the card right?

[radiance]

we did a test today between a system with 2x Quadro FX3800 in 2x PCI-E slots straight on the motherboard. (Asus WS)
we also did a test on a xpander box with 2x Quadro FX3800 both shares by an 8x PCI-E slot. (eg 2 cards on one 8x bus)

we did'nt notice any difference in speed between both.
it just took a few seconds longer to initially load the model/scene into memory after clicking the meshnode.

Radiance

Correct me if I'm wrong, but that would make perfect sense since the information is loaded into the VRAM, and so there's very little that has to be done outside the card, the only time there would be info transferring to and from, would be when there are initial mesh's being loaded onto the card, and if you had a lot of UV Maps or images that had to be loaded also, it could take longer yet, but once the info is loaded on the card, there's not a lot it has to do, off the card right?

no, once loaded it just sends tiny bits of information/commands.

there is no difference between 2 GPUS on 1 8x slot vs 2GPUs on 2 16x slots.
afaik that means 4x per GPU.

we did tests and the speed and response is identical.
it does take a little longer to load the mesh initially indeed.

Radiance

Radiance