OTOY Forums

azen wrote:

darkline wrote:Can anyone send me an email address and I can send you the file please?

Sent you a contact address via PM. I recommend WeTransfer for sending large scenes.

Done, thanks very much

Hi, just a quick follow up but how did you get on with my scene?

Did you experience the same issues as me? Can this be fixed, or does V3 really take up 2.5x the Vram of the same scenes?

Hi,

This is currently being treated as a standalone-based issue, as it is difficult to account for the extra memory consumption from all the usual plugin-related causes. The standalone is currently being investigated. Hopefully a forthcoming release will provide a resolution, or at least provide more understanding. Apologies for the inconvenience.

darkline wrote:Hi, just a quick follow up but how did you get on with my scene?

Did you experience the same issues as me? Can this be fixed, or does V3 really take up 2.5x the Vram of the same scenes?

Sorry, but we didn't receive a scene, could you send it again to me or azen. Thank you.

Hi Abstrax,

think I replied to the other thread, but I resent the file and got a download confirmation.

Just wondering if you were able to reproduce the problem and or see a possible fix?

thanks

darkline wrote:Hi Abstrax,

think I replied to the other thread, but I resent the file and got a download confirmation.

Just wondering if you were able to reproduce the problem and or see a possible fix?

thanks

Yes, we've got the scene and I had a look at it, but must have forgotten to get back to you and let you know of the conclusions. Apologies for that.

The issue is the modified acceleration structure we introduced with version 2, which needs more memory, i.e. the problem also exists in version 2. The modified acceleration structure renders faster than in v1, but requires more memory. Usually the memory consumption doesn't grow as much as in your example, but if you have a high triangle / volume density then it becomes significant. In your example the fur is the killer. It's super dense (something like 24 million triangles or so) and thus requires a lot of memory.

If you can render it in version 2 but not in version 3, try reducing the parallel samples value to 1, which should reduce the amount of memory that is used to store the samples state.

In a nutshell: It's not a bug but a limitation of the faster ray tracing acceleration structure introduced with v2, for which there is no easy "fix" or workaround.

Thank you for the explanation.

V2 had a poly limit of I think 19 million, so I reduced the fur by 50% to get it to work In that, and tried some displacement maps to bridge the gap. As you say the problem I have now is those scenes don't open in v3, even with parallel samples set to 1.

So looks like I'm stuck with v2 if v3 increases ram usage even more so. I just bought a 980ti but perhaps I need a 1080 for anything fur related. I've also just upgraded from 32gb to 64gb ram which I'm sure will help, so I'll try v3 again when the max plugin gets a more stable release.

Don't take this the wrong way but if octane is going in the direction of more ram usage I think it's something you really need to look at. If I can't render my single fur teddy bear, how's anyone going to use octane for making multiple characters with fur? I think it's a very serious limitation which will struggle being adopted by big studios because of this, and with each release it seems to increase ram usage, not optimise it.

Perhaps this is a limitation of all GPU renderes and it's simply a case of picking the right renderer for the job. But thanks for the explanation and I know what I can and can't do with it now. I'm really looking forward to adaptive sampling in 3.1.

darkline wrote:Thank you for the explanation.

V2 had a poly limit of I think 19 million, so I reduced the fur by 50% to get it to work In that, and tried some displacement maps to bridge the gap. As you say the problem I have now is those scenes don't open in v3, even with parallel samples set to 1.

The memory usage of the geometry in v2 and v3 should be the same. It's the sample buffer that required additional memory in v3. But you can reduce the memory usage of the sample buffer by setting parallel samples to 1. v3 doesn't have the film buffer on the GPU anymore which save a bit, but that becomes only significant if the the resolution is not tiny (as in your example scene). If that is not enough, then I assume that was a really tight fit on v2.

So looks like I'm stuck with v2 if v3 increases ram usage even more so. I just bought a 980ti but perhaps I need a 1080 for anything fur related. I've also just upgraded from 32gb to 64gb ram which I'm sure will help, so I'll try v3 again when the max plugin gets a more stable release.

Don't take this the wrong way but if octane is going in the direction of more ram usage I think it's something you really need to look at. If I can't render my single fur teddy bear, how's anyone going to use octane for making multiple characters with fur? I think it's a very serious limitation which will struggle being adopted by big studios because of this, and with each release it seems to increase ram usage, not optimise it.

Perhaps this is a limitation of all GPU renderes and it's simply a case of picking the right renderer for the job. But thanks for the explanation and I know what I can and can't do with it now. I'm really looking forward to adaptive sampling in 3.1.

If you want to do anything fur related, you would usually use the hair primitive which uses a lot less memory. Unfortunately there doesn't seem to be a way to fetch the actual hair information from 3ds max hair&fur, which is the actual problem here.

Thanks Abstrax

It may also be that the max plugin for v 3 is still undergoing development. I've seen reports of users getting blank screens with old v2 scenes. So I'll have to wait and see in that one.

Interesting about the hair primitive. Does this work with any other hair plugin? I'm wondering for example if using Ornatrix hair would use your hair features? Would this plugin be better and what kind of memory saving could I expect to see if so? (Don't want to do all the work again if it's minimal)

Thanks again!