Most visualization tools put a layer between your data and the image on screen: GLSL, a C++ render engine, a driver-specific path, often spread across several languages and libraries. Makie has always worked to keep that stack coherent, and this work takes it further.
After a quick tour of what makes Makie compelling today, I'll get into the new material: raytracing in Makie, what it makes possible for 3D data, volumes, and lighting, and where it's going, including hardware RT core support.
The heart of the talk is the new Vulkan backend, built on Lava.jl and set to replace GLMakie. The key idea is shaders written purely in Julia, compiled to SPIR-V. Because Vulkan covers general compute as well as rendering, the same approach extends beyond graphics, and it interoperates cleanly with CUDA.jl and AMDGPU.jl (ROCm) so you can mix it with the GPU code you already have. All acceleration happens in one language, cross-vendor across AMD, NVIDIA, and Intel, with zero-copy data flow from your arrays straight to the GPU. And because shaders are now just Julia, you extend the renderer the way you extend any Julia library: overload a method on your own type and it runs on the GPU. Custom materials, new primitives, and your own lighting all become high-level Julia callbacks.
I'll close with a look at where this is heading for Makie and Julia-native GPU computing more broadly.