Apex GPU: Run CUDA Apps on AMD GPUs Without Recompilation (github.com)

Apex GPU has launched a groundbreaking lightweight translation layer that allows NVIDIA CUDA applications to run seamlessly on AMD GPUs without the need for recompilation or source code modifications. Users can simply set an environment variable, and the tool will dynamically intercept CUDA calls at runtime, translating them to their AMD equivalents. This innovation makes it easier for developers and organizations to leverage the cost advantages and performance capabilities of AMD GPUs while maintaining compatibility with existing CUDA applications. This development is significant for the AI/ML community, as it provides a straightforward solution to a long-standing barrier that restricts CUDA's use to NVIDIA hardware. The APEX GPU supports 38 core CUDA functions essential for memory management, asynchronous operations, and kernel execution, as well as additional functionality for high-performance math and neural network operations. Such compatibility enables effortless deployment of machine learning models and data processing workflows on more economical AMD platforms, ultimately driving greater hardware diversity and cost-effectiveness in computational environments.