GPU Memory Full - will it be addressed in future updates?

Same issue with DVR 16 and DVR15
on Nvidia GTX 2x1080 8GB Each, 64GB RAM
on iMac with Radeon 4GB with 40GB RAM

10 sec timeline, Deflicker and Stabilizer enabled, no color corrections

Anyone else able to try and confirm this works for them also? When I get these errors:

DeviceResourceTypeCUDA.cpp:177: CUDA error cudaErrorMemoryAllocation (2)
or
cudaMallocArray failed: Requested size 5867.9 MiB (26894x14299-4 datatypesize 4

Toggling the 'Use GPU for Blackmagic RAE decode' saving then playing again makes it go away. It appears not to apply to what I set it to just that changing it stops these errors for a while; however, they do eventually come back.

I am running a live log tracer called CM Trace on resolvedebugger.txt so I can watch it as I work and as per the screenshot you can see it.

Not a complete fix but if confirmed by others is a easy way to work around it.


---


Forget all the BS above - jumped the gun on this one...

You did not give any information on your system configuration, so you'll need to be the one to review the link below and see if it's relevant. viewtopic.php?f=21&t=102152

Did you actually get a "GPU Memory Full" error displayed, or some other behavior?

Further to my post above, I figured out what was causing the issues. For me it was all about clip size coming out of fusion. Being new to this I did not realize that the BG layer sets the comp size within fusion, I was using the large JPG (6500x4000) to allow for motion tracking of PTZ and having come from an AE background, had to wrap my head around the differences. Anyway, Fusion trying to output a comp of this size was exceeding the memory capacity of the GPU and it ran out of space - still think this is a bug as it should not try to request more memory than available. Since fixing this I have not had another issue and while I can not be sure if this situation is the cause for anyone else I wanted to share just in case.

... Being new to this I did not realize that the BG layer sets the comp size within fusion, I was using the large JPG (6500x4000) to allow for motion tracking of PTZ and having come from an AE background, had to wrap my head around the differences. Anyway, Fusion trying to output a comp of this size was exceeding the memory capacity of the GPU and it ran out of space - still think this is a bug as it should not try to request more memory than available. Since fixing this I have not had another issue and while I can not be sure if this situation is the cause for anyone else I wanted to share just in case.

Agreed !!
I ran into the same Problem with an extremely simple 3D fusion animation of Planet Earth. Just a sphere, two lights and a little fog.
I mapped a texture of the earth surface to the sphere (size 8192x4096 4,35 MB) and an image of the milky way to the ImagePlane3D (size 8192x4096 1.81 MB)
Even with a timeline of FHD (1920x1080) the System runs into GPU Memory Full.
Just opening the project allocates about 6GB of GPU memory

Something has to happen for sure.
Maybe optional texture compression could help?

Resolve was just updated. Have not installed yet, however, wondering if anyone has seen any improvements?

Resolve was just updated. Have not installed yet, however, wondering if anyone has seen any improvements?

In addition to some other serious issues which have been fixed in the latest update 16.1.2 it seemes that BM has worked on this topic too.
The project which crashed immediately before is now rendering completely. The GPU memory usage is leveled slightly to the maximum, in my case between 6 and 7,95GB.

GPU Memory.png (59.01 KiB) Viewed 2852 times

So there was definitely a big lessons learned

Thank you BlackMagic Development Team

EDIT:
But it is extremely slow, the rendering took about 19 minutes for a simple 10 second animation

My laptop has both an integrated graphics card and a separate dedicated graphics card. I right clicked my desktop Resolve icon which gave me the option of opening Resolve with the integrated or the dedicated. Clicked on dedicated and problem seems to have gone for the time being.