VM crashing when faced with non-light browser load

I’m running bare metal TrueNASScale, and running several VMs on it. I am not passing it a GPU. The VMs are stable if I leave them be. If I access them via spice they are responsive and render everything quickly until I load a large image or other heavy multimedia element in a browser - Then they crash within seconds. A lot of sites have elements large enough to cause this.

I have tried firefox, Chromium and epiphany in Devuan. I have also tried Firefox in Debian and Mint. The behavior is the same - although epiphany + Devuan was slightly less frequently doing it. I have tried with and without swap. They have 4GB RAM assigned, and this is not filled at the time of the crashes, and assigning 20GB doesn’t change the behavior anyway. The rendered fps of mouse movement drops rapidly in the moments before the crash.

Quickest way to reproduce it is to load a 1080p youtube video - within 5 seconds it will crash. At 480p it plays fine indefinitely.

I have not been able to trigger this with large videos or images outside the browser.

Is this a known issue with the Scale implementation of KVM? Is there a fix? Is there an easy way in Scale way to give the qxl driver more VRAM?

Tried on a Windows VM and can’t replicate it. The Windows VM thinks it has 4GB of VRAM (or half the assigned RAM as shared VRAM), the Linux VMs think they have 64MB.

I really want to avoid trying to fix this with API calls to increase the available VRAM in Linux. But does anyone know what the calls would be? Or a better way to change them? Or know that it wouldn’t help?

Bumping this because I’m still having this issue, and was about to wrote this post again forgetting I’d already posted it.

Is there a GUI sanctioned way to increase VRAM (the presumed culprit)? Is there something I can do client side? Some sort of magic software rendering driver?