Go is experimenting with portable SIMD

Go’s SIMD experiment lets one instruction process several values at once, useful for loops over numeric data. The Go team describes a portable API for Go 1.27 alongside architecture-specific operations. Vector width stays out of the portable types, allowing implementations to choose hardware instructions or emulation. That portability does not promise equal performance on every machine.

The Go Blog · Sep 24, 2026

Why it matters

Worth following if you maintain CPU-heavy Go code and currently depend on assembly or external libraries.

More context

The experiment requires GOEXPERIMENT=simd. Try a small hot loop with the documented toolchain and benchmark representative data on each target CPU; the portable API is experimental.

Keep in mind

This is the Go team’s design and implementation report, not a 1min benchmark. APIs and supported operations can change.

Original read