ID 474156 - VRAM usage with GPU accelerated nodes gradually increases/leaks and there is no way to clear it in the Nuke session

Problem summary:
VRAM usage with GPU accelerated nodes gradually increases/leaks and there is no way to clear it in the Nuke session

This happens with ZDefocus and Convolve nodes at least, probably other GPU accelerated nodes too

Customer reported version:
nuke.12.2v6

Customer reported platform:
centos7

Steps to reproduce:

1) Open Nuke and Task Manager/a Terminal

2) Create Checkerboard node and connect a ZDefocus node and view it

3) In the ZDefocus node, make a change to the knobs, like the size knob in increments of 10 with maximum knob set to 1000, and wait for the Viewer to fully update

4) Keep making changes, and watch the VRAM usage gradually go up, either in Task Manager or with nvidia-smi | grep nuke replacing nuke for the Nuke path shown in nvidia-smi

5) Try clearing all the caches, or clearing the Nuke session, the VRAM usage does not go down

Expected behavior:
For the VRAM to not got up rapidly in the first place, and to be clearable

Actual behavior"
The VRAM usage gradually increased and could not be cleared

Workaround:
Close Nuke

Reproduced by Support in:
Nuke 16.0v4 - Windows 11
Nuke 13.0v1 - Windows 10, Centos 7.5
Nuke 12.2v1 - Windows 10
NukeX12.1v1 - Windows 10, Centos 7.5 - regression
 

Unable to reproduce bug in:
Nuke 12.0v6 - Windows 10, Centos 7.5
 

Earliest version tested
Nuke 12.0v6 - This issue doesn't appear in this version and has regressed