CUDA/OPEN CL ERROR

[hmercado]

Hi,
I am getting this CUDA/OPEN CL ERROR
Any clue what may be causing this? what does it mean. The CL one is printing OUT_OF_RESOURCES.
Does that mean my GPUs are getting overloaded with such a simple scene?

CUDA error screenshot
CL error screenshot

Thanks

(Edited by Alex to link screenshots externally)

[Alex]

Do you know your graphics card specs? Model and memory.

[hmercado]

Yes they are a evga Titan X Hybrid and a evga Titan Z. Each one has 12GB GDDR5

When I got the CUDA ERROR I was using the Titan Z - GPU(1) .

I then switched to the Titan X and got the same CUDA ERROR and then tried the Titan X with OPEN CL and got the CL error.

[Alex]

Hi again.

Definitively the graphics card power is not the issue.
Try to simulate the domains separately to see if the problem persists.

If it only happens when mixing the liquid and rigid domains, we can arrange some pack and go scene to troubleshoot the issue on our side.

Thank you

[hmercado]

Yes, it was only crashing when mixing the liquid and rigid domains.

I ran both simulations separately and there were no issues.

However I did some more tests. I removed the hub and connected the nodes directly removing from the rigid domain the daemons that were intended to affect only the liquid domain, which were affecting the rigid domain when they were connected to the hub and I did not get any more CUDA errors.

But I haven't tested the daemons individually to see which one could have been causing the crash and I did not test on OPEN CL yet.

I also noticed that in the example scene that you sent me showing the use of the rigid domain, the domains were not connected to a hub but directly to each other. Maybe that has something to do?

I also only connected the liquid domain to the rigid domain. This shows a one way direction as indicated by the arrow in the wire. But in the example scene the wire shows a two way direction between the two domains. Does that makes any difference to the relation?

Here's an img showing how the relation editor looks compared to the previous screenshots:
CUDA error fixed screenshot

(Edited by Alex to link screenshot externally)

[Alex]

Hi again,

I'm glad you managed to make some progress.

When your scene starts becoming a bit complex it is indeed a good idea to stop connecting everything through a hub to have precise control about what's actually linked.

When you press Tab on the Relationship Editor you can disable the default behaviour of connecting every new node to a hub.

Being told that, yes, the direction of the links in Dyverso domains matter.
You can try create an empty scene, add two domains with their emitters facing each other and experiment with the possible combinations.

So, you should try to connect again the rigid to the liquid domain and see if the problem arises again.
There is also a button in the relationship editor to show the implicit links (links created because of hub and group links).

Thanks for sharing your progress

[hmercado]

Hi,

Just to let you know, I Had the CUDA error occasionally with the final test scenes. That is a couple of time it crashed half way or so, but then other times it finish the simulation without any problems, without making any changes to the scene.

Here are the links to the final rendered tests:
https://youtu.be/1KQbmakLqvc
https://youtu.be/7aGI4ZMi-QU

Thanks again for all your help.

[Alex]

Thank you very much for the follow up. Both tests are really cool.

We are currently working on an update that hopefully may fix some of the random CUDA errors. We'll let you know when available.