I'm glad Intel has been swallowing their pride on this one, I want them to succeed for once.
What really would drive me to support them directly would be (para)virtualization support via SR-IOV/GVT-g. Especially given Microsoft's recent stunts an increasing amount of people (including me and close friends) are evaluating moving to Linux while keeping a Windows VM around, but both NVIDIA and AMD seem adamant about keeping those features gated to enterprise cards (even thought it's a software thing, see community efforts to reverse this[1]) so they can charge ludicrous prices to businesses for those.
My opinion is that if Intel wants to get their foot in the door they should win over enthusiast by breaking patterns such as this one rather than try to compete exclusively on performance.
I tried to hold out for the Intel launch anticipating sr-iov support, but I ended up refreshing hardware before. Any half-decent motherboard has their iommu groups proper, allowing passing any pci device into a vm with minimal overhead.
It sure sucks to need two cards, but I went/want AMD for Linux and still(sadly) NVIDIA for Windows.
It works really great. A 3 display setup where left is always Linux, right is always windows (when it's running) and middle is switched between inputs with ddcutil (cli tool to reconfigure your display).
I tried looking-glass framebuffer copy over a shared memory bus but it wasn't quite as smooth for me as switching inputs.
I'm on low-end gear, AMD 5600x, AMD 390, NVIDIA 3060, ASUS X570-P board. No tuning whatsoever, I let the Linux scheduler deal with it and it's good enough.
I can't recommend looking-glass enough as your spice client for mouse and keyboard capture, it's great.
Though this can all go fck right off if someone(Intel) does SR-IOV on consumer gear, I will replace my cards in a beat.
Problem is, there is no decent AMD card (and I will not touch nvidia for linux) with decent number of outputs for that use-case (in my opinion).
The cheapest ones are like 8 years old and out of support and still costs like $250, comes with a fan and still doesn't have decent outputs. And also takes a lot of PCI-E slots. A dual graphics card system means you barely have expandability options left after you have the hardware required to boot. And that ~one PCI-E slot still visible might be disabled if you decide to use another NVME-drive.
The amount of research required to pull it off is insane.
This is my biggest hope for Intel/ARC. A low powered card with decent number of outputs for a fair price and decent linux support.
What exactly are you looking for? There's both the single slot 6400 with dual outputs or the dual slot 6600 with 4 outputs. And both are modern RDNA2 architectures.
3+ digital outputs, preferably passive and preferably one slot (though many compromises have to be made).
6600 costs about $350 where I live, which is quite steep considering it's only purpose is to serve 2D and very basic 3D editing. Especially considering you'd also need a more capable/expensive card for the VM (at least for my usecase) .
Compared to A380 with 4 digital outputs which costs less than $200, it has the potential to be an extremely attractive linux card.
Have you considered a DisplayPort hub? MST support has been a thing for a long time, it should work pretty well (particularly since this is basically how ~every thunderbolt & usbc dock works)
I have MST on my displays and it is a disaster. Also, considering the usecase where you use the VM to drive one of the displays which breaks the MST chain.
For this reason and how displayport/gfx-vendors handle unplugged cables I do prefer HDMI/DVI. Wish it wasn't the case.
Maybe a MST-hub could work well but I've heard of some of the difficulties and pecularities a simple KVM-switch can suffer from so I don't have high hopes for it.
Their GPUs should also work really well in linux, since you don't really need DX11/DX9. All games with those graphic APIs get translated to Vulkan by [DXVK](https://github.com/doitsujin/dxvk) anyways.
You don't really need DX10 and lower on Windows anymore either, dgVoodoo2[1] is routinely used (along DXVK) on Windows itself because of how better supported DX11/DX12 and Vulkan are by modern drivers.
I hope someone will benchmark a game with DX9+dgVoodoo2 vs native DX9.
Off-topic, but if you're wanting to move to Linux and are worried about gaming, you should try running Windows games on Linux with Valve's Proton. I was worried too, but I got a Steam Deck and it's absolutely crazy how many games just work. I have generally mainstream tastes and over half of my Steam library is at least rated "playable". And there's rarely any tinkering you have to do beside click play (that seems to be one of Valve's sticking points for marking things playable, as if you had to do extra setup outside Steam it would be difficult to do on the Deck with no keyboard attached).
Most of the ones that don't are because of anti-cheat, which wouldn't work inside a VM anyway.
(On-topic, I want a consumer level card with SR-IOV too. I should be able to have my desktop be a VM in a closet without buying a license and card that cost more than a car.)
> What really would drive me to support them directly would be (para)virtualization support via SR-IOV/GVT-g.
Linus asked the Intel reps for that on the WAN show and the answer was that Intel was also going to have that he an enterprise only feature at least initially but they'd try to escalate.
But I wouldn't get your hopes up. I'd love to have it and it's really dumb it's so locked up, but nobody seems to want to risk devaluing the enterprise markup.
What really would drive me to support them directly would be (para)virtualization support via SR-IOV/GVT-g. Especially given Microsoft's recent stunts an increasing amount of people (including me and close friends) are evaluating moving to Linux while keeping a Windows VM around, but both NVIDIA and AMD seem adamant about keeping those features gated to enterprise cards (even thought it's a software thing, see community efforts to reverse this[1]) so they can charge ludicrous prices to businesses for those.
My opinion is that if Intel wants to get their foot in the door they should win over enthusiast by breaking patterns such as this one rather than try to compete exclusively on performance.
[1]: https://github.com/DualCoder/vgpu_unlock