OTOY Forums

Hello
I recently replaced a 780 Ti with a Titan X to be able to manage VRAM capacity.

On medium-size scenes, both cards work together fine in both the Live viewer and Final render.
But on a big scenes, I get this error message :

CUDA error 2 on device 1: The API was unable to allocate enough memory to perform the requested operation.
-> failed to launch kernel (info)
CUDA device 1: Failed to info channel for info pass
failed to finish info pass "Render layer ID"

But when I deactivate the 780 Ti, it works fine, but just slower

I am currently on 2.21.1 (following Oaktar's advice to avoid upgrading in the middle of a project)

Nvidia Pilot is up to date : 352.86

Could anyone help me to fix that and optimize my render settings ?

thx in advance

Chris

Sorry : here is my hardware :

X99E-WS socket 2011-3
Titan X + GTX 780 Ti
Xeon E5-3GHZ
64 Go RAM DDR4,
4 PCIe 16x

Sorry, but don't get what You don't understand here 780ti is 3GB while TitanX is 12 GB card,
when You load scene that takes more than smaller card can handle it fails.. - it's normal =)

Or I don't get something else here? =)

this is why i get a 780 (6Gb) next to my Titan instead of 780Ti (3Gb) while it is faster than 780. otherwise i would lose 3gig of Titan on dual card render process.

Sorry for my brain shortcuts on the matter

What you say is that what causes the failure is the low 3 Vram of the 780 Ti ?
I thought octane could "organize" power ressources from GPU !

To push this further, it means that if I have a big scene and use a PC with a 3GVRAM 780 Ti over Octane Render Network with stronger cards,
it is also bound to fail ?

doesnt matter network render neither single pc. it matters what the lowest amount of Vram. bcos octane works parallell so, sends the same data into each gpu(vram) that are active. Lets realize your scene needs 4Gb but the lowest vram of your gpus is 3Gb. then your render fails. you have to disable the lowest gpu or you can use "Out of Core" feature under Octane setting that helps to render but slower. Also helps you only on the textures side not the geometry. BUT, first should try to optimize the scene.

I have that same motherboard. If you run either all Maxwell cards or all Kepler cards it seems to run a lot smoother in general. It's kind of a touchy board with getting multiple GPUs to post correctly not run into errors. I've got x5 980s on it right now and it is smooth as can be.

You could try Out-Of-Core feature to save some texture memory. Tweak the scene (subdivision, instancing ...) on the geoometry.

Thank U all for replies.
I get it better now.
This particular project scene is a good "case study" for my short experience with Octane.
The model is a Race bike which is loaded with tons of small parts that I have to render in a shot.
No matter what the geometry hiding tricks or simplification, I came short on VRAM with my 780 Ti.... and rushed for a Titan X that could do the render on its own...
If I had done better research on GPU / VRAM issues with Octane, it would have saved me €€€ !

Keep up the good stuff

Chris