Nvidia’s Next Driver Makes Even More GPUs ‘Open’, in a Specific and Original Way

The headline of Nvidia’s latest GPU announcement should be read slowly, analyzing each clause as it comes in.

“Nvidia is going to make a full transition” sounds like a real commitment, a call to action. “Towards an open source GPU,” yes, evoking the company’s “first step” announcement a little over two years ago, so this must be progress, right? But let’s backtrack a word here, and then conclude with “GPU kernel modules.”

So Nvidia has “achieved equivalent or better application performance with our open source GPU kernel modules” and added new features to them. And now, most of Nvidia’s modern GPUs will default to open source GPU kernel modules, starting with the R560 driver version, which is dual-licensed under the GPL and MIT. But Nvidia has moved most of its proprietary features into a blob of closed-source proprietary firmware. The parts of Nvidia’s GPUs that interact with the broader Linux system are open, but the drivers and user-space firmware are none of your business or that of the OSS community.

Is it better than what came before? Certainly. AMD and Intel have maintained open source GPU drivers, both in kernel and user space, for years, but also with proprietary firmware. This brings Nvidia a little closer to the Linux community and allows for debugging and community contribution. There is no indication that Nvidia intends to go further in its open source efforts, however, and its modules remain outside the main kernel, packaged for users to install themselves.

Not all GPUs will be able to use the open source drivers: a number of chips in the Maxwell, Pascal, and Volta families; GPUs in the Turing, Ampere, Ada Lovelace, and Hopper architectures are recommended to switch to open bits; and Grace Hopper and Blackwell units must do so.

As noted by Hector Martin, a developer of the Asahi Linux distribution, at the time of the first announcementThis change makes it easier to sandbox closed source code when using Nvidia hardware. But the net amount of closed source code is about the same as before.

Nvidia’s blog post has details on how to integrate its open kernel modules on various systems, including CUDA configurations.