Plugins + Scripts VAMCompanion

[Case]

Most of it works on the RX 9070 XT, but as VR-user I could run into VRAM-shortages ... I'll write, if I have found good settings ... Only thing which still is only CPU is XTTS ... But there should be a solution (but maybe no totally easy one at the moment. It is only a matter of development time, though, people are working on it).

On both my mobile APU and the RX 9070 XT there was nothing much at release but it gets better every day ...

Yeah, I'm a VR user, too & a bit wary about the 16Gb VRAM on the 9070XT. It's a sweet gaming card, but my guess would be that in order to run VaM in VR AND run GPU-accelerated nontrivial local models, 24-32Gb would be basically mandatory?

Maybe your setup is simply the best for Radeon users - don't try to make the 9070XT do something that it's not optimized for (and doesn't have the software ecosystem for), and just outsource to a networked secondary PC/laptop's CPU (Probably also easier to get networks to run on CPU?)

Only bummer for me is that I'd need another SSD to set up my old PC - I didn't think buy a spare one before the storage madness started ...

 

[mmn123]

Yeah, I'm a VR user, too & a bit wary about the 16Gb VRAM on the 9070XT. It's a sweet gaming card, but my guess would be that in order to run VaM in VR AND run GPU-accelerated nontrivial local models, 24-32Gb would be basically mandatory?

Maybe your setup is simply the best for Radeon users - don't try to make the 9070XT do something that it's not optimized for (and doesn't have the software ecosystem for), and just outsource to a networked secondary PC/laptop's CPU (Probably also easier to get networks to run on CPU?)

Only bummer for me is that I'd need another SSD to set up my old PC - I didn't think buy a spare one before the storage madness started ...

Yeah, I can feel the pain, buying a ssd now ... There is one thing I found out: that the 16 GB VRAM were maxed out at first because of some strange AI integration in AMD Adrenalin (which loaded a LLM into the VRAM). Now it doesn't seem to do it anymore. Meaning: Maybe with better control of VRAM usage one could load a small one for inference in VAMcompanion. But it needs further experimentation ...