00:52 kode54: gotcha
00:53 kode54: I've heard the dxvk experience is mostly okay but vkd3d is mostly failure
01:01 DemiMarie: fililip: whether they should support it is only of theoretical interest unless you can persuade them that a device reset is not a driver bug
01:02 DemiMarie: or a hardware bug (fixable or otherwise)
01:19 DemiMarie: fililip: supporting robustness makes APIs more complicated (you can’t treat textures as immutable) and in the absence of eGPUs is not needed on hardware/drivers with good fault containment (Intel/Nvidia).
01:49 kode54: so convince amd to fix their fault issues
01:50 kode54: and at the same time, convince intel to fix their vulkan stack
01:50 kode54: or I guess I should just give up and buy a 4000 series nvidia card
01:57 DemiMarie: kode54: I’m just the messenger here
01:57 DemiMarie: None of the vendors will discuss their future roadmaps in public, presumably for legal reasons.
01:58 DemiMarie: What I can say is that my issue for robustness in Mutter is on their roadmap.
01:58 DemiMarie: So eventually Mutter should be able to recover.
02:00 DemiMarie: Process isolation would also help, and it’s necessary for security if you are doing virtualization with less than fully trusted guests using the GPU, or if you are implementing WebGL and/or WebGPU “correctly” (read: by having the sandboxed renderer process ship GPU machine code to the less-sandboxed GPU process, rather than shipping WebGL and/or GPU commands, thus trading off pordability for security)
10:11 fililip: i did a few experiments with shader timeouts and on my 6600xt i can never get it to hard reset (it always soft recovers), except when i just undervolt it too much for instance
10:12 fililip: but i think this is new behavior, before it could hard reset with shaders (and definitely does on strix due to vcn timeouts)