✅ Updated Nov 2026
Affiliate Disclosure: As an Amazon Associate, we earn from qualifying purchases. Links marked "Check on Amazon" are affiliate links — learn more.

Every Linux gamer with an NVIDIA card installs the same driver package, yet since 2022 that package has contained two different kernel drivers: the long-standing proprietary modules and an open-source set that NVIDIA itself now recommends for most current cards. The names sound like a licensing footnote, but the choice decides which GPUs are supported at all, how your distribution packages the driver, and how long an older card keeps receiving a working stack.

This explainer keeps to what NVIDIA documents: what the company actually opened, what deliberately stayed closed, which generations each flavor supports, and why the 560 driver series flipped the recommendation. It is a status map, not a benchmark comparison, and it ends with a simple rule for picking the right stack for the card you already own.

As an Amazon Associate, we earn from qualifying purchases.

Quick answer

NVIDIA publishes two kernel-module flavors for Linux. The open modules, dual-licensed since driver 515, cover Turing and newer GPUs and are NVIDIA’s recommended choice since the 560 series; Blackwell cards require them. Maxwell, Pascal and Volta cards use the proprietary modules only. Either way, the user-space driver and the GSP firmware remain closed source.

What “open” actually means here

In May 2022, alongside the R515 driver, NVIDIA published the source code for its GPU kernel modules under a dual MIT and GPL license, developed in the open on a public repository. That was a genuine change in kind, not in marketing: the kernel-space code that talks to the hardware, manages memory and schedules work can be read, audited and patched by anyone, and distribution maintainers can build it with the same tooling they use for the rest of the kernel.

The scope matters as much as the license. “Open” applies to the kernel modules only. NVIDIA did not open the whole driver, and the company has been consistent about that boundary from the first announcement. Think of the stack as three layers: the kernel modules at the bottom, the GPU System Processor firmware in the middle, and the user-space libraries on top. Only the bottom layer changed hands. Everything a game actually calls, the Vulkan and OpenGL drivers, the CUDA libraries, the video encode and decode components, ships exactly as it did before, as closed binaries from NVIDIA.

One architectural detail explains the support limits below. The open modules lean on the GPU System Processor, a RISC-V processor on the card that took over jobs the kernel driver used to do on the host CPU. That offload design is what made the kernel code practical to publish, and it is also why the open modules cannot simply be back-ported to every old card: the GSP exists only on Turing-generation hardware and later.

What stays closed

Three things remain proprietary no matter which kernel flavor you install. First, the user-space driver: the Vulkan, OpenGL, EGL and CUDA libraries that games and Proton actually load are closed-source NVIDIA binaries, identical for both module flavors. Second, the GSP firmware itself: the open modules load a signed firmware image onto the GPU’s processor, and that image is a binary blob, not source. Third, the supporting tools and feature stacks around the driver, such as the CUDA toolkit components, follow their own closed licensing.

The practical consequence is that “open” here does not mean a fully open graphics stack in the way AMD and Intel users understand the term. Their drivers are open from the kernel through Mesa in user space. NVIDIA’s arrangement is a hybrid: an open kernel foundation with a closed house built on top. That still buys real things, including kernel-integration work, public bug tracking on the module code and a cleaner story for distributions that dislike building closed kernel code, but it does not make the driver community-maintained, and fixes for game-facing bugs still arrive on NVIDIA’s schedule.

Which GPUs each stack supports

NVIDIA’s own documentation draws the support map by architecture, and it is asymmetric in both directions.

Blackwell cards, the GeForce RTX 50 series, are open-only. The proprietary kernel modules do not support them, so an RTX 5070 or RTX 5060 owner has no choice to make: the open flavor is the driver. Turing, Ampere and Ada Lovelace cards, meaning the RTX 20, GTX 16, RTX 30 and RTX 40 series, support both flavors, and since the 560 series NVIDIA’s documented recommendation for these generations is the open modules. Maxwell, Pascal and Volta cards, the GTX 900, GTX 10 series and the Titan V generation, are proprietary-only: the open modules require the GSP those architectures lack, so a GTX 1080 or GTX 970 owner stays on the closed modules for as long as their driver branch lasts. Cards older than Maxwell sit on legacy branches that no longer receive feature updates at all.

Datacenter GPUs were actually the first class NVIDIA declared production-ready on the open modules back in 2022, while GeForce support was initially labelled alpha quality. That ordering has since inverted in importance for gamers: the open modules matured through the 535, 545 and 550 series, and by the 560 series NVIDIA stated they had reached feature parity with the proprietary modules on supported GPUs.

Why the 560 series changed the default

The 560 release in 2024 is the hinge point of this story. With it, NVIDIA announced that the open kernel modules had achieved what it describes as full feature parity and equivalent performance on supported hardware, and that the company was transitioning fully to the open flavor as the recommended path. The driver installer began defaulting to the open modules on GPUs that support them, with the proprietary modules still available as an explicit choice for the generations that can use either.

Distributions followed the same direction. Mainstream packages now commonly offer the open flavor as the default or the recommended variant for Turing-and-newer hardware: Arch’s open-module package is the standard recommendation for those cards, Ubuntu offers open variants of its driver packages, and Fedora’s community packaging provides both flavors. For a gamer this mostly removes the decision: install the current driver the normal way for your distribution and, on a supported card, you get the open modules unless you deliberately override the choice.

Open modules are not Nouveau

One confusion is worth clearing up, because the names collide. Nouveau is the long-running community driver for NVIDIA hardware, built by independent developers, paired with the NVK Vulkan driver in Mesa. NVIDIA’s open kernel modules are a different thing entirely: they are NVIDIA’s own code, written and maintained by the company, and they pair with NVIDIA’s closed user-space driver, not with Mesa. A system running the open modules is running NVIDIA’s official stack with one layer’s source published. A system running Nouveau and NVK is running a community stack that NVIDIA does not ship. Both are legitimate choices with different trade-offs in features and support, but installing one is not installing the other, and advice written for one rarely applies to the other.

The cards in the reference table

The five cards below span the two situations a buyer can land in: Blackwell cards where the open modules are mandatory, and Ada Lovelace cards where both flavors exist and the open one is recommended.

GIGABYTE GeForce RTX 5060 WINDFORCE OC 8G

GIGABYTE GeForce RTX 5060 WINDFORCE OC 8G

GIGABYTE GeForce RTX 5060 WINDFORCE OC 8G

Check price on Amazon

  • Architecture: Blackwell
  • Memory: 8 GB GDDR7
  • Model: GV-N5060WF2OC-8GD
  • Kernel modules: open flavor required

GIGABYTE GeForce RTX 5060 Ti WINDFORCE 16G

GIGABYTE GeForce RTX 5060 Ti WINDFORCE 16G

GIGABYTE GeForce RTX 5060 Ti WINDFORCE 16G

Check price on Amazon

  • Architecture: Blackwell
  • Memory: 16 GB GDDR7
  • Model: GV-N506TWF2-16GD
  • Kernel modules: open flavor required

ASUS Prime GeForce RTX 5070 OC Edition

ASUS Prime GeForce RTX 5070 OC Edition find alternatives

ASUS Prime GeForce RTX 5070 OC Edition

Check price on Amazon find alternatives

  • Architecture: Blackwell
  • Memory: 12 GB GDDR7
  • Model: PRIME-RTX5070-O12G
  • Kernel modules: open flavor required

ASUS Dual GeForce RTX 4070 EVO OC Edition

ASUS Dual GeForce RTX 4070 EVO OC Edition

ASUS Dual GeForce RTX 4070 EVO OC Edition

Check price on Amazon

  • Architecture: Ada Lovelace
  • Memory: 12 GB GDDR6X
  • Model: DUAL-RTX4070-O12G-EVO
  • Kernel modules: either flavor, open recommended

MSI GeForce RTX 4060 Ventus 2X Black 8G OC

MSI GeForce RTX 4060 Ventus 2X Black 8G OC

MSI GeForce RTX 4060 Ventus 2X Black 8G OC

Check price on Amazon

  • Architecture: Ada Lovelace
  • Memory: 8 GB GDDR6
  • Model: RTX 4060 VENTUS 2X BLACK 8G OC
  • Kernel modules: either flavor, open recommended

How to choose

Choose by the card’s architecture, not by preference. If your card is Blackwell, install the open modules, because they are the only supported option. If your card is Turing, Ampere or Ada Lovelace, follow NVIDIA’s documented recommendation and take the open modules, which is also what current distribution packages will hand you by default. If your card is Maxwell, Pascal or Volta, install the proprietary modules and accept that your card lives on whatever driver branch still supports it. In every case, prefer your distribution’s packaged driver over a manual installer download, since the packages handle kernel rebuilds and Secure Boot signing for you, and both flavors need signed modules on a Secure Boot system.

Who should skip this

Skip the whole question if you game on AMD or Intel graphics, where the open Mesa stack makes the choice for you. Skip it if your NVIDIA card is older than Maxwell, because neither current flavor applies to legacy-branch hardware. And skip the manual flavor override if your distribution’s default already matches your card’s generation: for most gamers the interesting decision ended with the 560 series, and the remaining work is simply keeping the packaged driver current.

FAQ

Are the open kernel modules the same as the Nouveau driver?

No. Nouveau is an independent community driver that pairs with Mesa. NVIDIA’s open modules are the company’s own kernel code, published under a dual MIT and GPL license, and they work with NVIDIA’s closed user-space driver. They are separate stacks with separate maintainers.

Is the whole NVIDIA driver open source now?

No. Only the kernel modules are open. The user-space Vulkan, OpenGL and CUDA libraries remain closed NVIDIA binaries, and the GSP firmware the modules load onto the GPU is a signed binary image rather than published source code.

My GTX 1080 runs the proprietary driver. Should I switch?

You cannot: Pascal cards lack the GPU System Processor the open modules require, so the proprietary modules are the only option for Maxwell, Pascal and Volta hardware. The open recommendation applies to Turing-generation cards and newer.

Do the open modules perform differently in games?

NVIDIA states that the open modules reached feature parity and equivalent performance with the proprietary modules on supported GPUs as of the 560 series, which is why it made them the default recommendation. Individual results always vary by game and system, so treat parity as the documented position rather than a promise about a specific title.

Final verdict

The open kernel modules are best understood as NVIDIA moving the foundation of its Linux driver into the open while keeping the game-facing layers closed. For Blackwell owners they are simply the driver. For Turing, Ampere and Ada owners they are the documented recommendation since the 560 series, delivered by default through normal distribution packages. For Maxwell, Pascal and Volta owners nothing changed. Check your card’s generation, install the matching flavor through your distribution, and the question is settled.

Explore Our Guides & Free Tools