Starlight Sharp - CUDA out of memory

Topaz Video Bugs and Issues
1

Starlight Sharp failed when processed 1280x720 video with upscaling to 1920x1080. Logs show CUDA out of memory. It is happen after processing about 1 minute of the source video.
Processor AMD Ryzen Threadripper 3970X 32-Core Processor 3.70 GHz
Installed RAM 128 GB (128 GB usable)
Storage 954 GB SSD Samsung SSD 970 PRO 1TB, 3.64 TB SSD WDC WDS400T1R0A-68A4W0, 954 GB SSD Samsung SSD 970 PRO 1TB, 224 GB SSD Kingston SHPM2280P2/240G, 3.64 TB SSD WDC WDS400T1R0A-68A4W0
Graphics Card NVIDIA GeForce RTX 3090 (24 GB) x2 SLI
Edition Windows 10 Pro
Version 22H2
Installed on ‎17.‎11.‎2022
OS Build 19045.6456

logsForSupport.zip (14.7 MB)

2

Try lowering the max memory in the Preferences Panel to 80% and see if that helps the GPU handle the models task on your file.

File>Preferences>Processing

  1. Ensure that your GPU is selected for the AI Processor

  2. Set the max memory to 80%

  3. Set max processes to 1

  4. Save the settings and close the panel.

  5. Re-run your clip through the process that you were trying.

1 Like
3

same here:
RTX 3090 / 24GB and Starlight Sharp exits with CUDA out of memory after processing 19sec. of Video 640x320 to 1920x960 .. so the output of several videos is only 19sec. long
I have set the max ram to 80% btw
System Ram is 32gb

my other RTX 5070Ti / 16GB is working fine with max memory set to 100%

4

Did you try setting the max. RAM to 70% or lower?

5

this night I tried max 70%
960p (3x) fails after 19 seconds again
640p (2x) is working