Wave – A universal GPU instruction set architecture
WAVE is a new universal GPU instruction set architecture designed to allow code to run on multiple GPU vendors without modification. It supports NVIDIA, AMD, Apple, and Intel GPUs, producing identical results across platforms. The architecture includes various components and has been verified on several hardware systems.
- ▪WAVE allows for a single binary to run on different GPU vendors including NVIDIA, AMD, Apple, and Intel.
- ▪The architecture features 11 hardware-invariant primitives and has passed 102 conformance tests.
- ▪Benchmarks show the Apple M4 Pro achieving 3,587 GFLOPS for F32 matrix multiplication.
Opening excerpt (first ~120 words) tap to expand
WAVE: Wide Architecture Virtual Encoding The ARM of GPU computing. One binary, any GPU. WAVE is a vendor-neutral GPU instruction set architecture. Write GPU code once, run it on NVIDIA, AMD, Apple, and Intel GPUs unchanged. The same binary produces identical results on all four vendors. ARM defines what a CPU is so multiple vendors can build compatible chips. WAVE does the same for GPU computation. Highlights 11 hardware-invariant primitives across 4 GPU vendors 34,000 lines of Rust across 10 crates 618+ unit tests, 102/102 conformance tests passing Verified on Apple M4 Pro, NVIDIA T4, and AMD MI300X 3,587 GFLOPS F32 matrix multiply on M4 Pro (53.5% of Apple MPS) 89.29% CIFAR-10 accuracy via PyTorch integration, matching native exactly Install pip install wave-gpu Or build from source:…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.