OTOY Forums

Octane, abandon Maxwell for Pascal.

Is the title irritating?

I had to struggle with an error that could not be handled in the last two weeks of re-rendering the file.

I had to fight with an unknown Cuda error while updating Octane and Plugin and re-sizing and rendering.

I never thought the Octane Render was a problem. So internally I thought I had a problem with my work or a hardware problem and found a solution.

So I wasted two weeks. It was a huge waste of time and money.

I used all the ways I could. I deleted the objects one by one, and changed the graphics cards one by one.

The problem was not solved at all.

And I thought about it. What has changed in the past work and growing the size .....

Change of some objects and revision of material.

And I got to know.

Updated version of Octane Render ... Updated .... Updated ... Updated ....

Yes. The problem was the program.

We were rendering on Maxwell TitanX using Octane 3.03.4 and Plugin 3.03.4.1.

So I switched from the main computer back to the previous version.

The main computer had an error every hour until this morning.

But now it is rendering without problems.

Does Octane 3.04.3 run on Pascal TitanX without problems?

I have rendered the same file in Pascal, but still have an unspecified error.

Of course, it does not work at all. 5 to 10 frames are rendered.

As a result, developers have yet to optimize for CUDA 8.0 GPUs.

I invested $ 5,000 and bought four TitanX.

This is a huge problem for the worker on the project.

It took me two weeks to figure out what the problem was. In the meantime, all the projects have been discontinued and the company has had to pay for it.

There is still a major problem with Pascal since version 3.04. I want you to think seriously.

Also, it is a big problem that developers can not test because the hardware is not ready.

Those who buy and use it right now are using the latest hardware and are working on the hardware based project.

I'm sorry, but after a lot of tests I've not been able to reproduce the problem here. I am not sure if the problem is related to Maxwell or Pascal or the combination of the two GPUs.

So only to be sure ... do you say that the scene is working fine in Maxwell, and not work in Pascal? ... sorry, I can't understand to you very well.

There is something really weird in this scene, but I can't figure what could be the problem. I've seen other reports about CUDA errors, but nothing like your problem. Here I don't have this combination of Pascal and Maxwell GPUs, and the scene is really huge to fit in my workstation, but I've made a lot of tests with the scene instances, and all seems to work fine.

The main question is ... can you render the scene in the Maxwell GPUs or not? And in the Pascal GPUs? ... it is only to know if the problem can be related to one or other GPUs type.

Thanks,
-Juanjo

juanjgon wrote:I'm sorry, but after a lot of tests I've not been able to reproduce the problem here. I am not sure if the problem is related to Maxwell or Pascal or the combination of the two GPUs.

So only to be sure ... do you say that the scene is working fine in Maxwell, and not work in Pascal? ... sorry, I can't understand to you very well.

There is something really weird in this scene, but I can't figure what could be the problem. I've seen other reports about CUDA errors, but nothing like your problem. Here I don't have this combination of Pascal and Maxwell GPUs, and the scene is really huge to fit in my workstation, but I've made a lot of tests with the scene instances, and all seems to work fine.

The main question is ... can you render the scene in the Maxwell GPUs or not? And in the Pascal GPUs? ... it is only to know if the problem can be related to one or other GPUs type.

Thanks,
-Juanjo

--------------------------------------------------------------------------------------------------------------------------------------
I found the exact cause of the problem, although it is not accurate.

In my "Point_Think_Small_Bush.lwo" file, the instance is applied.

This creates an instance in the Surface Type.

I applied the Offset value here, and there are some data that apply the alignment value of the Rotate value as the Normal value.

If the object on which the instance is created is flat, there will be no problem, but if this is a bend, then suppose that the Offset value, the Normal direction, and many instance entities are entangled and error.

I'm still testing it, but I think that if this instance becomes more than a certain amount in a very heavy work file, something else will cause problems and lead to errors.

So I'm testing it with the offset value subtracted.

As I opened the Scene file to test whether this problem occurred even in 11.6 version, all job file instance values were initialized and I rendered it.

It works without error as expected.

So, only the current offset is initialized and is being rendered.

I suppose this is a problem in version 3.04.3

I will know if I wake up tomorrow.

juanjgon wrote:I'm sorry, but after a lot of tests I've not been able to reproduce the problem here. I am not sure if the problem is related to Maxwell or Pascal or the combination of the two GPUs.

So only to be sure ... do you say that the scene is working fine in Maxwell, and not work in Pascal? ... sorry, I can't understand to you very well.

There is something really weird in this scene, but I can't figure what could be the problem. I've seen other reports about CUDA errors, but nothing like your problem. Here I don't have this combination of Pascal and Maxwell GPUs, and the scene is really huge to fit in my workstation, but I've made a lot of tests with the scene instances, and all seems to work fine.

The main question is ... can you render the scene in the Maxwell GPUs or not? And in the Pascal GPUs? ... it is only to know if the problem can be related to one or other GPUs type.

Thanks,
-Juanjo

----------------------------------------------------------------------------------------------------------------------------------------------

As you might expect, this is a problem that appears when you apply an offset to an Instance Type while applying a Surface. It appears after version 3.04. There is no error in 10 hour test.

juanjgon wrote:I'm sorry, but after a lot of tests I've not been able to reproduce the problem here. I am not sure if the problem is related to Maxwell or Pascal or the combination of the two GPUs.

So only to be sure ... do you say that the scene is working fine in Maxwell, and not work in Pascal? ... sorry, I can't understand to you very well.

There is something really weird in this scene, but I can't figure what could be the problem. I've seen other reports about CUDA errors, but nothing like your problem. Here I don't have this combination of Pascal and Maxwell GPUs, and the scene is really huge to fit in my workstation, but I've made a lot of tests with the scene instances, and all seems to work fine.

The main question is ... can you render the scene in the Maxwell GPUs or not? And in the Pascal GPUs? ... it is only to know if the problem can be related to one or other GPUs type.

Thanks,
-Juanjo

----------------------------------------------------

I am still testing today.
I know one more thing.
This means that when the Offset value of the Instance or the Rotate Alignment value is Normal, an error occurs due to collision with something else in the Scene.

I have not yet found what the instance value of the Scene collides with. You can confirm that Offset and Rotate Alignment values are set to Normal.

I thought it was a problem for Instace, but it seems to be a complex problem.

Maxwell TitanX + version 3.03.4 is good

Maxwell TitanX + version 3.04.1 is an error

Pascal TitanX + version 3.04.1 works fine "Offset or Rotate Values off"

Thanks for your tests. I'm trying to reproduce the problem here isolating the instancers that you have found with problems, but without luck yet.

Thanks,
-Juanjo