Nvidia Updates CUDA Software Policy

What’s going on here?

Nvidia has updated its End User License Agreement (EULA) for the CUDA software to explicitly ban the use of translation layers for running CUDA-based software on non-Nvidia hardware platforms. Previously, this information was only specified in the online version of the EULA, but starting with CUDA 11.6, it is now also included in the EULA file that accompanies the software installation. This move aims to prevent other companies, including notable ones like Intel and AMD, as well as Chinese GPU manufacturers, from using CUDA code on their devices through translation methods. The update was noticed by a software engineer known as Longhorn, which brings attention to the stricter enforcement by Nvidia of its licensing terms.

What does this mean?

The inclusion of the new clause in the installed CUDA EULA documentation means that Nvidia is taking a stronger stance against the use of its CUDA software on non-Nvidia hardware. This decision could have significant implications for projects like ZLUDA, which aim to enable CUDA compatibility on Intel GPUs, and for various Chinese GPU makers who have shown interest in running CUDA code. The restriction makes it clear that only the recompilation method (available to the developers of the respective CUDA programs) is now considered a legitimate way to run these programs on non-Nvidia hardware, directly impacting the ability of competitors to utilize Nvidia’s CUDA architecture without engaging in a direct partnership or developing their solutions.

Why should I care?

If you’re involved in software development, especially in fields heavily reliant on GPU acceleration such as AI, machine learning, or graphical computations, Nvidia’s stricter licensing terms could influence your choice of hardware and software for future projects. The ban on translation layers means that running CUDA applications on non-Nvidia hardware could become more challenging, potentially limiting the flexibility to choose hardware based on cost, performance, or availability. Moreover, for those interested in the broader tech ecosystem, this move by Nvidia could signal how major tech companies intend to protect their proprietary technologies from being used on competitors’ platforms, which could lead to more isolated tech ecosystems or provoke innovation in response.

For more information, check out the original article here.