[HN Gopher] GPU Video acceleration in the Windows Subsystem for ...
___________________________________________________________________
GPU Video acceleration in the Windows Subsystem for Linux now
available
Author : Fudgel
Score : 37 points
Date : 2023-02-13 21:48 UTC (1 hours ago)
(HTM) web link (devblogs.microsoft.com)
(TXT) w3m dump (devblogs.microsoft.com)
| dharma1 wrote:
| For GPU accelerated ML in Win11 (PyTorch w/ CUDA) - is WSL enough
| nowadays, or still dual boot to Linux?
| sxp wrote:
| I was able to get Stable Diffusion and other similar ML systems
| working in WSL2 on Win11. There might be performance
| differences between WSL and a native system. I haven't
| benchmarked AI in different configs, but my main use case is
| Rust+wgpu and there were noticeable performance differences
| because the GPU driver exposed by WSL didn't have as many
| features as the GPU driver used on Windows. I also had problems
| with other APIs on WSL such as Optix.
| Fervicus wrote:
| I want to be able to use IntelliJ from WSL. Can I do that yet?
| shmerl wrote:
| I'd prefer MS to stop pushing DX12 NIH for a change and start
| using Vulkan.
| gjsman-1000 wrote:
| NIH? Vulkan was released nearly a year _after_ DX12. And DX12
| follows a long line of DirectX versions.
| kevingadd wrote:
| What's stopping Intel, AMD and NVIDIA from offering native
| Vulkan support in WSL, then?
|
| This 'NIH' (not so as already explained by another commenter)
| is offering graphics acceleration to WSL guests based on
| existing Windows drivers, which compensates for the fact that
| the GPU vendors aren't already offering acceleration - Vulkan
| or otherwise - for WSL guests.
|
| I'm actually not sure how I would get stable acceleration,
| Vulkan or otherwise, in a Linux host in any VM. In my
| experience acceleration in VMWare and VirtualBox are both a
| crapshoot to the point of not being worth using.
| bitwize wrote:
| For 25 years now the standard for 3D graphics has been DirectX.
| Why would you expect Microsoft, of all companies, to support
| anything else?
| Femtiono wrote:
| You are aware of the history of DX right?
|
| Their SDK is old and really good.
|
| Just because Vulkan exist doesn't mean DX is invalid.
|
| Good to have more than one thing. Innovation and stuff
| kevingadd wrote:
| Arguably without the OpenGL vs D3D and now Vulkan vs D3D back
| and forths, along with experimental APIs like MANTLE, we
| definitely wouldn't have a lot of the robust tech we have
| access to today.
|
| OpenGL's freeform experimentation and evolution with
| extensions let people test things out in production
| environments to figure out what worked, while D3D's stable
| feature set meant that games and productivity software could
| - if it made sense for the developer - choose to ship a more
| limited feature set that worked _everywhere_ , all of the
| time.
|
| D3D also has consistently offered great debugging tools and a
| robust reference rasterizer, things you simply can't get in
| an OpenGL environment. As a game developer it's invaluable to
| be able to swap over to a Direct3D backend for debugging even
| if you end up using OpenGL as your default. (These days,
| Vulkan has first-class debugging support too, which is
| great.)
|
| Now we have Vulkan as the new home for experimentation and it
| has great debugging and validation layers, while D3D pushes
| forward on certain new features and provides a more
| consistent baseline on Windows desktops. For console games as
| well, you can use Vulkan on (AFAIK) Nintendo Switch, while
| using D3D12 on Xbox, so each API is providing value for
| console game devs as well.
___________________________________________________________________
(page generated 2023-02-13 23:00 UTC)