Mesh shaders on RDNA graphics cards
gpuopen.com
gpuopen.com
That is funny considering that Vulkan has added support for mesh shading roughly a year ago [1]. And, if you take a look at the members list of the Khronos group, you will find that AMD is listed in the top category [2] (what ever Ikea is doing there).
[1]: https://www.khronos.org/blog/mesh-shading-for-vulkan [2]: https://www.khronos.org/members/list
They probably have custom renderers for rendering the flat/cartoon illustrations for the assembly instructions, and PBR renders for rendering realistic views.
Grab a bunch of of these resources and put them in a box and you have living room design without having to manufacture, assemble, light, and shoot it in the real world.
eta: ope, I see someone beat me to the same comment.
* "Multiverse of madness - how we render millions of doors": https://www.youtube.com/watch?v=HXEVncxtmxg
Product visualisation seems to be a bit of a recurring theme for Blender talks over at least the past couple of years.
He could have solved most of his issues by switching to a 3d configurator. I have built a custom threejs engine over the last decade that allows you to declaratively define materials, texture, repeating geometry, ... In a few lines of code. It has primitives like flexible grids, cloning, CSG operations etc.
I currently use it mostly for outdoor living structures, but implementing doors and/or floors for this would be trivial.
If you'd do this using my engine, the biggest thing left would be gathering the requirements and conventions... So I'd build a configurator for door configurators that allows suppliers to easily create their own configurators...
In fact, I'm doing this right now for terrace roofs/sun rooms.
Turtles all the way down...
It's much easier to digitally assemble those models into a virtual kitchen scene than to build an actual kitchen. Especially things like a sink where you'd need to cut a custom countertop to mount it.
The simple answer that you can infer from the video is that they make very high quality models for the design stage, which is definitely cheaper than making prototypes in all plausible material combinations.
Then you can use those models for the simple product pictures too, and packing a 3d model with generic clothes is probably faster than having a prop department do it.
IKEA most likely cares about WebGL and WebGPU, which in any case is irrelvant for Mesh Shaders as they are decades away to arrive in the browsers if past performance is any indication of Web 3D adopting native 3D features.
There doesn't seem to be anything inherently insecure about mesh shaders and they now have working examples in both Vulkan and Metal, so I'd expect them to show up sooner rather than later.
I want Vulkan to be great, but it's painfully obvious it's behind.
This is exactly what happened with mesh shaders - nVidia shipped the first hardware, but the extensions could rightfully not be uplifted to a standard API before other vendors also delivered implementations, as that could inconvenience their efforts.
It has been the same following DirectX lead, from GPU vendors collaboration with Microsoft, and extensions spaghetti as OpenGL.
Additionally it seems GLSL is stagnant, and from 2023's panel the suggestion was to adopt HLSL instead as there is no budget for language research.
http://web.archive.org/web/20140627040243/http://www.cgsocie...
Also, RIP CGSociety. Just went looking for that article and found they shut down earlier this month.