CUDA error 719 on device *: unspecified launch failure Tweet

[nuno1980]

@teeshiiddo:
What are your 2 videocards instabled (device 3 and 4 as error 719)? Have you any EVGA GTX 780 Ti Classified (my same issue - HERE)?

[zoppo]

I have the same issue from time to time, definitely not a hardware problem.

It only happened after I had updated from 3.0.4.5 to 3.0.5.1

Could you try downgrading to 3.04.5 again and test again? If the problem goes away, could you then upgrade to 3.05.x again and check if the problem comes back? If it does, could you send us the scene plus some instructions how to break CUDA?

I am in the middle of two projects don't have the time atm, sorry. It would be quite a hassle too, because my render slaves are not connected to the internet by default. So I have to switch them back and forth when installing a new OC version.
Maybe at the weekend.

Yesterday I encountered another strange crash: I see all 3 clients in the C4D picture viewer . The onscreen info shows that one or two of them are not rendering, but the clients' standalone windows do not show an error. When checking via GPU-Z I see that indeed the graphic cards are idle.

[teeshiiddo]

@teeshiiddo:
What are your 2 videocards instabled (device 3 and 4 as error 719)? Have you any EVGA GTX 780 Ti Classified (my same issue - HERE)?

I have 5x 780ti, 2x are OC version, the others standard (3x EVGA, 2x Nvidia).

I truncated the log on the first page, eventually all the devices failed with the same code. How do you isolate a device number to each GPU? All use Hynix memory.

[teeshiiddo]

I have the same issue from time to time, definitely not a hardware problem.

It only happened after I had updated from 3.0.4.5 to 3.0.5.1

Could you try downgrading to 3.04.5 again and test again? If the problem goes away, could you then upgrade to 3.05.x again and check if the problem comes back? If it does, could you send us the scene plus some instructions how to break CUDA?

For me at least I don't think the issue is version related. It has been on going 4-5 months.

I would be great if Otoy could find a way to recover from these errors automatically. Like I mentioned in an earlier post I now use Deadline render manager to restart C4D command line when it detects a render error. Sometimes it will report 20 errors in 12 hours, other times 0. It is very frustrating.

One plus point I'm still about to work in C4D GUI with octane while Deadline renders in the background. It would be great if Octane supported multiple instances natively. Also allowing Deadline two instances per node, meaning C4D could mesh one frame while octane renders another. Meshing/preparing often takes longer than the render.

[nuno1980]

I have 5x 780ti, 2x are OC version, the others standard (3x EVGA, 2x Nvidia).

I truncated the log on the first page, eventually all the devices failed with the same code. How do you isolate a device number to each GPU? All use Hynix memory.

Devices 3 and 4 = your 2 videocards with "EVGA OC version"? If "yes" then try to disable these 2.
What's EVGA OC version? SC, FTW or Classified...?

Hynix memory issue. yeah.

[aoktar]

One plus point I'm still about to work in C4D GUI with octane while Deadline renders in the background. It would be great if Octane supported multiple instances natively. Also allowing Deadline two instances per node, meaning C4D could mesh one frame while octane renders another. Meshing/preparing often takes longer than the render.

You can run multi-C4D's and run one octane on each one.

[jang2235]

Are you using data from a version earlier than 3.04.3?

Is not the data you created in the old version?

[zoppo]

This morning I installed 3.05.03 and up to now no crash - no CUDA error, no "silent" crashing where the SA window showed no error although the cards weren't rendering.
For testing purposes I rendered two scenes that did crash one or two of the three clients - no crash with 3.05.03 (*knocks on wood*).