Bend is a massively parallel, high-level programming language designed to automatically scale computations across many cores without explicit parallelism annotations.
Bend addresses the challenge of writing code that naturally exploits parallel hardware by providing a language with Python and Haskell-like expressiveness—including higher-order functions, closures, unrestricted recursion, and continuations—while automatically distributing work across available processors. The language is powered by the HVM2 runtime and can target GPUs, CPUs, or sequential interpreters. Code written in Bend will parallelize automatically when its structure permits; the same program can be executed sequentially or in parallel simply by changing the runtime target, with near-linear acceleration based on core count when run on massively parallel hardware.
Bend suits projects where you want high-level language features combined with the ability to harness many-core systems without managing threads, locks, or synchronization primitives. It is particularly valuable for computations with inherent parallelism that would otherwise require manual thread management or GPU-specific frameworks. The tool currently excels at scaling with core count rather than single-core performance, and it targets NVIDIA GPUs specifically. Windows support is still in development, though WSL2 provides a workaround.
The project shows active development with ongoing work to improve code generation and optimization techniques. Performance enhancements are expected as the implementation matures. The team is actively addressing platform support gaps and expanding GPU compatibility beyond current NVIDIA-only support.
