Share your success (visually, practically) with sys-gui-gpu

General Discussion
1

I’m relatively new to Qubes still, 6 months or so in, but through help on this forum, I’m now able to do everything I set out to do to eliminate my daily Debian system. Even some customized qrpc services, (carefully) driven by udev in sys-usb

One thing I’ve not yet tried is sys-gui-gpu

I’m interested in sys-gui-gpu as my understanding is that it can reduce/eliminate use of the CPU for graphics rendering

Like some other users that I’ve noticed on the forum, it’s not entirely clear to me what the UX is like, visually and procedurally when sys-gui-gpu is set up. And how many users are having success, and if they had to change their workflow

Specifically, I’m not understanding what is the workflow like for using 3-4 different AppVMs concurrently, and what it looks like. Would it be seamless, equivalent to using CPU to render in dom0 if set up properly?

Is there a thread or documentation that shows these things explicitly that someone can point me to? I’m not looking for docs on setting it up, rather I’d like to see it. Despite being technical, I’m a visual person, and this concept is challenging to grasp

What I’d really like to see before even attempting it is what it actually looks like, and whether it would fit my needs.

Is sys-gui-gpu the same UX as using a “standard” setup, or are there constraints that would hinder my patterns of use? Which are: having a dvm for browsing, an appvm for development, and one or two additional appvms that I interact with while using the other two.

Last, my system has a BMC with VGA output (currently unused), an i915-based PCI card (Intel ARC, used for dom0 output) and two NVidia GPUs (used for compute, so, not relevant). Could/would the BMC VGA serve any useful purpose should I use sys-gui-gpu, if I was using the Intel GPU for the sys-gui-gpu appvm? For example, could the BMC VGA act as a concurrently running fallback, to interact with dom0 if the sys-gui-gpu setup becomes “broken”?

I’m sorry for being yet another “what is sys-gui” post, and I must say thank you in advance for anyone willing to post something visual from their sys-gui-gpu configuration, or an explanation of what their workflow is like with sys-gui-gpu, to help me to understand if it is worth attempting.

2

No, the only purpose is to detach the GPU from dom0 to reduce attack surface further. It does not help rendering the content of qubes.

3

If it is really a BMC - which means you talking about an enterprise level server hardware (please confirm) - then no.
That is usually a completely separate computer board (whith it’s own CPU, RAM, and VGA, USB, EThernet connections) just inside the same ‘box’ and it’s purpose to (remotely) manage and configure the main hardware itself.

So , the main OS has nothing to do with that VGA output.

4

Yeah, it’s a real BMC. It comes integrated with WRX90 boards

I’ve only ever worked with server BMC interfaces over IP (the virtual KVM junk) so seeing that this had a VGA port made me curious. It would take some magic, I guess, to render the real framebuffer onto that VGA output. “Magic”, meaning I see what you’re saying, it probably doesn’t work like I imagined

Thanks

5

Ah, sorry, I had read that several times before but was confused by sys-gui vs sys-gui-gpu, thinking that sys-gui did not offload rendering, but sys-gui-gpu did

Apologies for having asked that question for the 1001st time on the forums

Thanks

6

No problem, I will try to improve the documentation