Search Issue Tracker
Fixed in 2020.2
[GPU PLM] VRAM not released following a GPU PLM bake
1.Download the attached repro project. Open using the Editor version above.
2. Ensure that we have enough texel density in Lighting Settings to require a juicy GPU bake
3. Ensure Progressive GPU is selected as lightmapper backend
4. Hit generate
Using a tool like GPU-Z it should be possible to see how much VRAM is being used by Unity during the bake. This is likely to be quite high as I have artificially pumped up the texel usage in my test project. If you've got the time, let the bake finish. Observe that once the bake has finished, Unity doe not release the VRAM used during the bake.
Once the bake has finished, the amount of VRAM used by Unity should return to somewhere around what Unity was using before the bake was started. The amount used does not just include the additional VRAM required by the lightmaps now added to the Scene. Seems somewhere more in line with the buffers assigned by our OpenCL kernels. This will be a lot on real-world Scenes. Especially considering we frequently run out of memory. Not a lot will be left for Windows and I have observed really poor Windows performance following GPULM.
All about bugs
View bugs we have successfully reproduced, and vote for the bugs you want to see fixed most urgently.
- Error in console when using the progress bar API along with Texture compression
- NullReferenceException when selecting a Dropdown Component override in the Prefab Overrides menu for the second time
- USS variables are broken on IOS
- Crash when project is reopened or upgraded
- Hub doesn't install windows 10 sdk properly