In this JuliaHub webinar, Dr. Tim Besard, Software Engineer and GPU lead at JuliaHub, gives a practical introduction to Metal.jl, the Julia package for programming the GPUs in Apple Silicon Macs.
He starts with high-level array programming and moves on to writing your own GPU kernels. He then covers what makes Apple GPUs different, and how new precompilation work brings first-kernel latency down from seconds to milliseconds in a fresh Julia Session.
What you’ll learn
Array programming on the GPU with MtlArray: broadcasting, reductions, linear algebra and FFTs
Writing, launching, profiling and debugging custom kernels with @metal
Writing portable kernels with KernelAbstractions.jl that also run on NVIDIA, AMD and Intel GPUs
Apple Silicon specifics: unified memory, storage modes and floating-point limits
Optimising kernels with SIMD groups
Precompiling GPU code for fast load times, and the Metal.jl roadmap
Who it’s for
Julia users with intermediate experience. No prior GPU programming experience is needed.
