[HELP] Can vsviewer be configured to be quicker?

[GlitterKill]

I believe I remember a thread somewhere that discussed why vs viewer takes so much longer to display vs the viewer that StaxRip uses, and believe it had  something to do with scanning the movie and/or building some sort of model from it before the filters are applied and the preview shows.  Is there a way to change the behavior so that models are not built before display?  

I am assuming (and may be totally incorrect; if so, I apologize), that this is for the various advanced cuda processing features that Hybrid supports.  What if a user only uses Hybrid for video filters, that do not leverage any cuda?  This is my case.  I use an AMD 9070 XT and do not have access to traditional cuda.  Is there any sort of argument or flag that I could start Hybrid with to change that preview behavior?  

On another note, I am hoping that someday ZLUDA may help AMD gpus run CUDA based video filters.    Are you familiar with it at all?

[Selur]

I believe I remember a thread somewhere that discussed why vs viewer takes so much longer to display vs the viewer that StaxRip uses, and believe it had  something to do with scanning the movie and/or building some sort of model from it before the filters are applied and the preview shows.

Whether index files are created depends on the source filter used.
Different source filters can be faster or slower building indexes depending on the source.
All source filters that create indexes will read through the whole file to create the index.
No clue what StaxRip uses for preview (maybe mpv?), but any meaningful speed differences are probably down to using a different or differently configured source filter.

Is there a way to change the behavior so that models are not built before display?

Use a different source filter.

Is there a way to change the behavior so that models are not built before display?

Interference of machine learning models only takes place if these are used and unless you use filters from the torch- or vsmlrt-add-on (or REALESRGAN) those are created since the filters are not used.  vsmlrt works fine with Intel and AMD cards, only the torch-add-on uses CUDA.

What if a user only uses Hybrid for video filters, that do not leverage any cuda?

Then no engine files or similar are created.

On another note, I am hoping that someday ZLUDA may help AMD gpus run CUDA based video filters.  Are you familiar with it at all

Yes, but last I checked it was nowhere near usable.

Cu Selur

[GlitterKill]

Thank you very much for answering my questions.