v1.28 render time

Thu Jan 30, 2014 8:55 pm

Bug?

v1.26 had an example render at 26 hrs to completion. v1.28 with the same file gets bumped up to 170 hrs. Is 1.28 letting us render longer, or is the 170 hours just wrong? The progress bar seems to suggest I really DO have that much time to render with.

As a side note, I created a separate post here for reporting a 1.28 bug without placing it in the 1.28 forum. Seems like everyone is reporting bugs there and if someone wants an answer to a bug post, someone like me might post a new bug before theirs gets answered.

Thu Jan 30, 2014 9:01 pm

treddie wrote:Bug?

v1.26 had an example render at 26 hrs to completion. v1.28 with the same file gets bumped up to 170 hrs. Is 1.28 letting us render longer, or is the 170 hours just wrong? The progress bar seems to suggest I really DO have that much time to render with.

As a side note, I created a separate post here for reporting a 1.28 bug without placing it in the 1.28 forum. Seems like everyone is reporting bugs there and if someone wants an answer to a bug post, someone like me might post a new bug before theirs gets answered.

Can you try exactly the same scene in 1.26 and 1.28? If those numbers still hold up, could you send us the scene so we can investigate?

Thanks,
Marcus

Thu Jan 30, 2014 11:57 pm

Let me play with it for a bit...I need to finish up this render tonight. When I end it, I'll do two more tests, one in .26 and the other in.28

Fri Jan 31, 2014 8:39 am

Seems like it has to do with heavy texture usage. When I'm pushing the limit of memory is when the problem seems to come up. If I dump a lot of the meshes and textures in my scene, the problem goes completely away and the times between 1.26 and 1.28 are identical.

See attached image of 1.26 vs. 1.28 times.

Incidentally, my other post about being able to render a scene with slightly moderate to heavy texture usage, only after a click on one of the individual textures or meshes, seems fixed in v1.28.

Fri Jan 31, 2014 8:54 am

treddie wrote:Seems like it has to do with heavy texture usage. When I'm pushing the limit of memory is when the problem seems to come up. If I dump a lot of the meshes and textures in my scene, the problem goes completely away and the times between 1.26 and 1.28 are identical.

See attached image of 1.26 vs. 1.28 times.

Speed Problem 2.jpg
Incidentally, my other post about being able to render a scene with slightly moderate to heavy texture usage, only after a click on one of the individual textures or meshes, seems fixed in v1.28.

Ok, I have an idea, which change have caused this issue, but I need a scene to reproduce the problem and verify a potential fix. Do you think you can create/derive a scene that shows the speed drop?

Just for completeness, could you also check, that the problem still exists in 1.29?

Thanks,
Marcus

Fri Jan 31, 2014 7:48 pm

If I zip up the package, it is still 274Mb...Too large for the site to accept as an attachment. Can you give me a temporary link to an upload area?

Sun Feb 02, 2014 6:21 pm

https://www.wetransfer.com/

Sun Feb 02, 2014 8:10 pm

Thanks remon_v. They set me up with Mega.co.nz. But I like to keep bookmarks to new services, so thanks for the link!

Sun Feb 02, 2014 8:18 pm

Ok, we had a look at the scene and we can't reproduce the slow-down here. Neither on Kepler nor on Fermi GPUs. Do you've got a second GPU and if yes, is SLI enabled in the NVIDIA driver settings? Which driver version are you using?

Thanks,
Marcus

Sun Feb 02, 2014 11:09 pm

GTX-260 Driver 9.18.13.3182
GTX560 Driver 9.18.13.3182

The 260 is relegated to handling my monitor, and the 560 is dedicated to rendering.

I do not have any option for SLI in my NVIDIA Control Panel nor through Device Manager's Display Adapters properties.

UPDATE:
OK, I went into the control panel after doing some Googling, and did discover that in place of "Set SLI Configuration", I have "Set PhysX Configuration", which I assume is an updated version? Anyway, I went in there and I only have a field called "Select PhysX processor" which I kept at "Auto-select". I then backed out of the PhysX area and opened up "Manage 3D settings". In there, I went to the octane app and forced "CUDA - GPUs" to only the 560 (It was set to "Use Global" which automatically had selected both GPUs). This SEEMED to have cleared up the problem. But then I thought I would set everything back to the way it was and in doing so, I could not recreate the problem. My feeling now is that I had not rebooted my system in days and that somewhere along the line something got messed up. When I rebooted yesterday, I think memory got cleared and the problem went away.