Wait, so how does it work right now? Could’ve sworn I’m using hardware H.264 with OBS Studio via FFmpeg VAAPI.
EDIT: Ah, for Vulkan. So what does that do?
If this works out, then we won’t need specific hardware dedicated to codecs. Buggy drivers can be bypassed as long as the GPU supports Vulkan. VAAPI could be eliminated since everything is moving to Vulkan.
We can use GPU acceleration on Nvidia without needing to use special hacks for their drivers.
VMs that can have Vulkan pass through will be able to have hardware enabled acceleration.
There’s quite a few ideas that this could unlock.
how iGPUs and dGPUs work in linux is something i find very confusing. would some kind soul please demystify this for me? how does this driver work for Radeon chipsets/dGPUs, and how does that affect iGPUs for systems with them?
its not too much different than on a windows system, you basically have a driver, and the driver basically has an internal whitelist on what hardware is supported on said driver, be dgpu or igpu. depending on distro/how you manage updates, your update schedule can vary.
the obly difference is on linux, theres usually a propietary driver option, and an open source option, where in windows, most people will just use the propietary closed sources option given by their chip manufacturer
no, i understand all of that, of course.
what i’m referring to is the whole hybrid mode/passthrough situation where dGPUs and iGPUs somehow depend on each other and dGPUs can’t really be run descretely.
oh that thats more a hardware question, which turns into a software one depending if hardware is present. basically if a manufacturer for laptops specifically deaigns a laptop that has both, they can choose to implement a hardware mux or not. the decision affects fps, latency and cost. if they lack it, software decides which ones to use and has to redirect dgpu to igpu, which is connected to the laptop screen (what causes latency and performance loss, as there is no hardware switch that goes from gpu to monitor directly)
however this odd phenomena where gpus get routed into igpu has a couple of benefits in very niche usecases and experiments. one example is that it allows for gpus that have no output to work as a typical gpu (e. g miner specific built gpus) and a joke example ia when a modder made a game run both DLSS 3.0 and FSR 3.0 frame generation by having a discrete nvidia gpu pass all its frames (including geneated ones) to a amd apu that appled amd frame gen as a propf of concept. this odd connection allows for interesting quirks not present if something is directly connected to the gpu.