CUDA Error 700... Tweet

[jang2235]

The Pascal Titan X Cuda 700 Error appears.
In the same environment, the Maxwell Titan X problem does not occur.

See send the log file.

The scene is a working file using a 10GB memory.

[juanjgon]

This week I'll release new plugin build using the new Octane 3.04 API. Perhaps these problems with the Pascal GPUs are fixed in the new Octane build ... if not, we can further investigate.

Thanks,
-Juanjo

[Lewis]

Yes I'm getting those more and more with new plugin releases, I hope OTOY will investigate this much better than so far.

Thanks Juan

[abstrax]

Yes I'm getting those more and more with new plugin releases, I hope OTOY will investigate this much better than so far.

Thanks Juan

We haven't received any scene yet that allowed us to reproduce the issue. If we can't reproduce it we can't fix it. So if you could send us a scene, we will try to reproduce and fix it as soon as possible.

[Lewis]

Yes I'm getting those more and more with new plugin releases, I hope OTOY will investigate this much better than so far.

Thanks Juan

We haven't received any scene yet that allowed us to reproduce the issue. If we can't reproduce it we can't fix it. So if you could send us a scene, we will try to reproduce and fix it as soon as possible.

Can you use LW scene or just Octane Standalone ?

[abstrax]

Yes I'm getting those more and more with new plugin releases, I hope OTOY will investigate this much better than so far.

Thanks Juan

We haven't received any scene yet that allowed us to reproduce the issue. If we can't reproduce it we can't fix it. So if you could send us a scene, we will try to reproduce and fix it as soon as possible.

Can you use LW scene or just Octane Standalone ?

I can try getting Lightwave running here in the office, which might take a couple days, depending on the Lightwave licensing and what we already have here in the office, which i would have to check first. If you can also reproduce it in the Standalone it would simplify and speed up things for us of course. Thanks for your help.

[juanjgon]

Lewis, if you have a scene to reproduce this problem, you can export it to Standalone from the IPR window and check if in Standalone you have the same problem. Anyway you can send the scene to me, and I can export it to Standalone here and make some tests.

-Juanjo

[Lewis]

I can try getting Lightwave running here in the office, which might take a couple days, depending on the Lightwave licensing and what we already have here in the office, which i would have to check first. If you can also reproduce it in the Standalone it would simplify and speed up things for us of course. Thanks for your help.

You don't need license for testing (you can use LW in Demo (Discovery) mode for Octane license). Yes I know it could be better in standalone but i have very minimal knowledge of Standalone GUI that.

But thing that bothers me is that this is not really scene related anymore (if it ever was), I had it on very simple stuff an don bigger scenes also but it's just random. If i could find 100% reproducible steps i would send it instantly to Juan.

And it's not hardware related 100% 'coz I'v even gone that far to try Bitcoin mining (with Pascal GPU enabled miner) on all my GPUs individually and all together + CPUs in same time just to stress machine 100% and see will it crash and it's stable 24/7. That' was my 2nd thing after i tested Heaven, Unigen and Furmark Benchmarks on My GPUs and all pass fine, just Octane crashes/problems.
Also Furmark and Bitcoin/Altcoin mining uses way more power out of GPUs and Temps are higher then when doing Octane rendering so it's pushing GPUs even more than octane 3.x can. Also this GPU mining Pascal support is base don Final CUD A8.0 SDK and 1080 smashes TitanX in the speed by 25% while in Octane my TitanX (M) is still few points ahead of 1080s which brings me to believe that Octane CUDA 8.0 support is still not up to level it could be so maybe that experimental Pascal support is what gives us this problems/errors.

Do you guys have any ETA on full/final CUDA 8.0 support in OR 3.x ?

Juan do you have Pascal GPU(s)now ?

Thanks

[juanjgon]

Juan do you have Pascal GPU(s)now ?

Not yet, I am waiting for a new Pascal laptop ... anyway I can add one to my Windows workstation to make some tests if nobody else can reproduce this problems from Standalone. Do you know if only one Pascal GPU is enough to reproduce it?

-Juanjo

[abstrax]

I can try getting Lightwave running here in the office, which might take a couple days, depending on the Lightwave licensing and what we already have here in the office, which i would have to check first. If you can also reproduce it in the Standalone it would simplify and speed up things for us of course. Thanks for your help.

You don't need license for testing (you can use LW in Demo (Discovery) mode for Octane license). Yes I know it could be better in standalone but i have very minimal knowledge of Standalone GUI that.

Ok thanks for the info. I guess we will get Lightwave working here one way ore another. So if you've could send me a scene that shows this error, even if it takes a while to occur, I'm happy to investigate.

But thing that bothers me is that this is not really scene related anymore (if it ever was), I had it on very simple stuff an don bigger scenes also but it's just random. If i could find 100% reproducible steps i would send it instantly to Juan.

And it's not hardware related 100% 'coz I'v even gone that far to try Bitcoin mining (with Pascal GPU enabled miner) on all my GPUs individually and all together + CPUs in same time just to stress machine 100% and see will it crash and it's stable 24/7. That' was my 2nd thing after i tested Heaven, Unigen and Furmark Benchmarks on My GPUs and all pass fine, just Octane crashes/problems.
Also Furmark and Bitcoin/Altcoin mining uses way more power out of GPUs and Temps are higher then when doing Octane rendering so it's pushing GPUs even more than octane 3.x can. Also this GPU mining Pascal support is base don Final CUD A8.0 SDK and 1080 smashes TitanX in the speed by 25% while in Octane my TitanX (M) is still few points ahead of 1080s which brings me to believe that Octane CUDA 8.0 support is still not up to level it could be so maybe that experimental Pascal support is what gives us this problems/errors.

Currently it doesn't look like we can improve things considerably for Pascal compared to what we have right now. Maybe a few percent, but not 25%. I think that's because Octane is quite complex compared to most other CUDA applications and mostly memory bound so it's mostly improvements in the memory area that help Octane and not so much improvements in computing performance. But I have spent only a limited amount of time on this, so I might be wrong here.

Do you guys have any ETA on full/final CUDA 8.0 support in OR 3.x ?

Switching from CUDA 8.0 RC to CUDA 8.0 didn't make any difference unfortunately. We still use CUDA 7.5 as part of the compilation pipeline, because otherwise things would be up to 50% slower... We are in contact with NVIDIA, but there is only so much they can do without accessing our kernel code. I think, eventually, I will have to have a proper look and check if there is anything that can be done to fix the problem with CUDA 8.

Juan do you have Pascal GPU(s)now ?

Thanks