WAVE launches universal GPU ISA across hardware vendors

WAVE attacks the GPU vendor lock-in problem from the bottom up by abstracting the 11 core primitives shared across 16 major microarchitectures. It is a critical leap toward write-once, run-anywhere machine learning workloads. Instead of relying on heavy APIs like OpenCL, WAVE compiles a compact binary that thin backends translate into native assembly. Built-in PyTorch integration allows developers to write custom kernels without maintaining fragmented CUDA and Metal codebases. The project highlights how converging GPU architectures across major vendors make a unified, hardware-invariant IR practically feasible. By providing native SDKs for Python, Rust, C++, and TypeScript, it massively lowers the barrier to cross-platform ML deployment.