As the title says my Nuke renders are going SLOOOOOOWWWWWWWWWWWWWWWWWWWWW on my render nodes.
If I load up the same machine, and just execute the same cmd line operation it takes <6 minutes. If I run it through deadline each render takes 50-60 minutes.
Same if I just open nuke and render it that way.
I can even run the exact same command line while Deadline is rendering and finish 10x faster.
Also the Deadline Nuke is running at 100% CPU. So I don’t know what it’s doing but something is super wrong.
Ok I just did the same thing to 5 machines and this is where things get weird.
If I run the cmd it runs plenty fast, normal fast.
If I run it as a Deadline Job it goes super slow.
If I change it from “Disable GPU” to “Enable GPU” in the nuke settings nothing happens.
If I put a GPU into the computer (old crappy GPUs) it runs at normal speed. Even before I install drivers and it’s just a “standard VGA device” so definitely no CUDA or OpenCL etc, just a GPU present physically in the machine… it renders normal.
I just did that with a machine.
Rendered slow. Physically installed GPU, started up machine. Rendered great (without drivers). But it renders fine without a GPU if I run it from the command prompt.
Hi Gavin,
Can you confirm what exact version of Nuke you are using? GPU rendering isn’t possible when running as a NT service. Are you running your Deadline Slave as a service? If so, that might explain why nothing happens. I just checked both submission scripts and plugin code and the “–gpu” flag is only set if you are using Nuke >= 7.0 and the checkbox is ticked, so can you make sure the “Use GPU For Rendering” option is un-checked. It’s off by default in our scripts, although it is sticky between sessions.
However, the above doesn’t explain the general slowdown you are seeing. Can you enable slave verbose logging in repo options > application logging and send us a full log report so we can see where the slowdown is? I know The Foundry are having a few issues with the latest version of Nuke…(if your running v8.0v5, it would probably be worth downgrading until a newer version is released). Also, if you send us a log report, it will contain the exact command line arguments which are executing on your machines, which I assume is the identical cmd line flags you are using, when you render the job via the cmd manually?
I’ll see if I can re-produce this today at my end as well for you.
Thanks,
Mike
Running Nuke8.0v4 still.
Command:
Nuke8.0.exe -V 2 --nukex -i -x -X "Write2" -F 35-39 "C:\Users\renderadmin\AppData\Local\Thinkbox\Deadline6\slave\RENDER-I7-14\jobsData\53b1d39fb5b7e13c54c163f1\thread0_tempd1jQv0\AMZN_BOATS_MASTER_A03.02.nk"Shot in the dark, but Deadline’s output handlers could be choking on the verbose output…
So far, I’ve been unable to reproduce this issue. Can I see a log report with timeStamps?
Do you see any difference if you remove the “-x” (taking it out of Batch mode in Deadline?)
How would I remove the -x from a deadline job? The command line always renders fine.
“-x” is Batch Mode, so you just need to un-check this option before submitting the Nuke job.
Feel free to send us the log report via the ticket system.
Thanks,
Mike
Hey Gavin,
As Mike alluded to earlier, it could be related to Deadline running as a service. Can you confirm whether or not your Deadline slaves that are rendering these Nuke jobs are running as a service? If they are, can you change one to run the Slave as a normal application instead of a service to see if that makes a difference?
Cheers,
Ryan