Apple Silicon Macs put a capable GPU within reach of Julia users. This webinar
introduces Metal.jl, which lets you accelerate numerical code and write custom
GPU kernels directly in Julia. We’ll start with high-level array programming
using MtlArray, covering broadcasting, reductions, linear algebra and FFTs,
before exploring custom kernels, debugging and profiling. We’ll also show how
KernelAbstractions.jl lets you write portable kernels for Apple GPUs and other
GPU backends.

The second half explores what makes GPU programming on Apple Silicon different,
including unified memory, storage modes and floating-point limitations. We’ll
then examine recent advances in kernel precompilation and caching, showing how
package authors can reduce first-kernel latency from seconds to milliseconds in
a fresh Julia session. Through code examples and pre-generated demonstrations,
attendees will learn how to get started, understand performance tradeoffs and
build Julia packages that make effective use of Apple GPUs.

Intermediate Julia knowledge is recommended; no prior GPU programming experience
is required.