Zluda 6

Zluda 6 is a compatibility layer that allows unmodified CUDA applications to run on non-Nvidia GPUs, including AMD and Intel hardware. It translates CUDA API calls and GPU kernels into the target platform's native instructions at runtime, enabling GPU compute workloads to execute without source code modifications or recompilation. This solves the vendor lock-in problem for developers and researchers who rely on CUDA-accelerated libraries but want to use alternative GPU hardware. Based on the official project blog, Zluda 6 introduces significant performance improvements and broader GPU support compared to earlier versions. The project is open-source and targets Linux systems, with experimental support for Windows. It is not a drop-in replacement for all CUDA features—some advanced capabilities like dynamic parallelism may have limited support—but it covers the majority of common CUDA workflows. The release marks a milestone in GPU compute portability, especially for AI/ML inference, scientific computing, and rendering tasks that depend on CUDA.

The official Zluda blog announced version 6 with major performance and compatibility improvements, generating discussion on Hacker News about GPU compute portability.