~8ms CPU & GPU overhead on the first running instance of the editor

I’ve been trying (and failing) to optimize a project that seems to have an overhead of ~8ms both in CPU and GPU time no matter what I do, while doing A/B editing on a secondary instance of the editor I noticed that if I loaded the exact same project, the secondary instance performs as expected. This behavior can be replicated using a Notch Block in TouchDesigner too. At first I thought it was a TD bug but the same behavior appears in Notch Builder (issue report @ TD forums). FXPlayerHost exhibits no such issue with a single running block.

Notch v0.9.23.248
NVIDIA RTX 4090, driver v528.49
Windows 10 22H2

How to replicate
Load the shared project file in 2 Builder instances, see them perform differently.

Being a new user I can’t attach the project file here, however, I have made it (temporarily) available here: test.dfx - Google Drive


There’s more at the TD thread.

Furthermore, this code runs poorly as a block in the latest FXPlayerHost bundled with Notch Builder, but it runs as expected in the one linked at the docs: Using FXPlayerHost to test Blocks | Notch Manual 0.9.23

Issue was due to NDI Sender inadvertently being enabled in the block (facepalm), disabling this leads to the expected performance. This, however, is not the case when using other media servers, TouchDesigner, for instance, takes less than 1ms to send a surface of the same resolution via NDI. Perhaps this is something to look at.