They'll do it if it makes the most sense for their profits, either through games or more likely AI computations. We just need AMD being competitive to push them in that direction. I'm open to purchasing an AMD card, but so far haven't found their performance compelling against NVidia.
I'd argue that they've only not been compelling for the last generation and a half. They've on balance delivered far better performance per dollar historically, and that only stopped with maxwell and their R 3 series.
There's also the proprietary NVidia architectures like hairworks and Physx. I just need the raw performance of an NVidia card. Maybe, back around 2003 I would have picked up an amd product.
Just to put something out here, Intel has officially abandoned the xf86-intel-graphics drivers.
Is there a decent driver for onboard graphics for Intel? I wouldn't really know as all the systems I currently run don't have onboard graphics. (that might change when the Razer tri screen laptop comes out)
Debian, Ubuntu and Fedora does not recommend the xf86-video-intel graphics driver. Quote from Phoronix from last year:
Now all relevant Intel hardware will be using xf86-video-modesetting, which accelerates 2D operations over OpenGL using the generic GLAMOR module -- the same approach used by the xf86-video-ati/xf86-video-amdgpu drivers for GCN and newer and optionally older hardware too.
With Fedora Rawhide as of today and obviously then beginning with Fedora 26, the switch is happening from xf86-video-intel to xf86-video-modesetting. Fedora already has been using xf86-video-modesetting for Skylake graphics while now the change-over is happening for all other Intel IGPs.
I have a question about the upcoming single card support you noted is going to be in the future updates to the 4.10 kernel. What sort of latency will be added with doing single gpu passthrough. I ask as I'd like to use vega/future amd cards in both linux and the windows VM as I have a vive. Latency however really matters to the vive so could be an issue if the single passthrough adds to much.
Does that mean it will be possible to use hardware acceleration on VMs even with a cpu that doesn't have VT-d, like my I7-3770K ?? Or am I illuded?
I believe you need vt-d for hardware pass through.
I guess you're right. I was just hoping that a new approach to virtualization could bypass this limitation.