QEMU 5.0 Kicks Off For Development

[phoronix]

Phoronix: QEMU 5.0 Kicks Off For Development

Following yesterday's release of QEMU 4.2, the next version of this open-source processor emulator for hardware virtualization entering development is QEMU 5.0...

http://www.phoronix.com/scan.php?page=news_item&px=QEMU-5.0-Opens-Development

[timofonic]

Any hopes of having vendor agnostic virtual GPU support?

[polarathene]

Any hopes of having vendor agnostic virtual GPU support?

The current ones are only for Intel aren't they? AMD and Nvidia can do hardware passthrough with VFIO, other than that there is the virtual graphic adapters(Bosch/VMWare/QXL/etc, can't quite recall them all), and recently the virtio-GPU which afaik only works with Linux atm for forwarding OpenGL to a host card to handle, that has some work towards Windows support and a Vulkan variant I think?

There's also SR-IOV, but that's for commercial grade GPUs not consumer level. There was some discussion in a past thread here between me and Bridgman about the possibility of AMD supporting the vGPU features that Intel offers. I don't remember specifics, but I don't think it was a no, just it wasn't likely to be something we'd see any time soon.

[Calinou]

Any hopes of having vendor agnostic virtual GPU support?

There's a recap of QEMU graphics devices here: https://www.kraxel.org/blog/2019/09/...vices-in-qemu/

[kpedersen]

Any hopes of having vendor agnostic virtual GPU support?

As part of my PhD, I have some initial work to provide an implementation of OpenGL (2.x) that works in an agnostic manner across a network and into a "client" such as a WebGL enabled browser or simple C/C++ viewer.
It is fairly effective because of the way OpenGL retains memory on the "GPU" and the rest of the messages are fairly small.

You can read some more info here (mostly about a new network game architecture rather than the GPU forwarding itself).

http://eprints.bournemouth.ac.uk/29501/1/ssa_short-ref5.pdf

or a slightly longer (https://tinyurl.com/s6c6kom)

Why this is fine for Qemu (and digital preservation) is that it is written entirely in pure ANSI C so is 100% portable (minus the network stack). It can even run quite happily on Plan 9 and MSDOS (with trumpet TCP/IP stack)

I am certainly planning on open-sourcing it upon completion but if it interests you (if you are happy with OpenGL 2) I am happy to share if you have a specific project or requirement in mind.

[AndyChow]

The current ones are only for Intel aren't they? AMD and Nvidia can do hardware passthrough with VFIO, other than that there is the virtual graphic adapters(Bosch/VMWare/QXL/etc, can't quite recall them all), and recently the virtio-GPU which afaik only works with Linux atm for forwarding OpenGL to a host card to handle, that has some work towards Windows support and a Vulkan variant I think?

There's also SR-IOV, but that's for commercial grade GPUs not consumer level. There was some discussion in a past thread here between me and Bridgman about the possibility of AMD supporting the vGPU features that Intel offers. I don't remember specifics, but I don't think it was a no, just it wasn't likely to be something we'd see any time soon.

There is virglrenderer which with virtio-GPU allows OpenGL acceleration on Windows guests. Works pretty well but unfortunately you have to compile qemu with virglrenderer enabled, which it isn't in most distro packages. Afaik there was work on supporting Vulkan, but not sure the current status.

[towo2099]

Where you see a windows driver for virio-GPU?
There is only a proof of concept driver, which is not production ready.

[polarathene]

There is virglrenderer which with virtio-GPU allows OpenGL acceleration on Windows guests. Works pretty well but unfortunately you have to compile qemu with virglrenderer enabled, which it isn't in most distro packages. Afaik there was work on supporting Vulkan, but not sure the current status.

So... basically what I said, minus recalling the name of what virtio-GPU was using under the hood? I don't follow the progress anymore, but last I heard Windows support wasn't ready. OpenGL usage in Windows is also low afaik unless you have some abstractions for the graphic APIs commonly used on the platform?

[AndyChow]

So... basically what I said, minus recalling the name of what virtio-GPU was using under the hood? I don't follow the progress anymore, but last I heard Windows support wasn't ready. OpenGL usage in Windows is also low afaik unless you have some abstractions for the graphic APIs commonly used on the platform?

Yes, plus the mention that the compilation of qemu had to be done manually for it to work. I remember getting it to work ok with windows maybe two years ago. The acceleration was there. I mostly abandoned it since dxvk and derivatives (d9vk) pretty much made gaming on linux a reality for most games, directly in steam with proton.

I couldn't get it to work with android, but when I ran android with sdl2 an gl=on I got gl acceleration. I think I used the qxl driver.

[BillFleming]

If using arch VIRGL just works out of the box. Been compiled in for a while now.