Red Hat to author new Linux driver for Nvidia GPUs in Rust

Original Article ↗ Hacker News Discussion ↗

Overall reaction to Red Hat’s Rust Nvidia driver

  • Many welcome the effort as good news for Linux users, especially for newer Nvidia GPUs.
  • Some see it as part of a broader “tide turning” toward Rust and safer drivers.
  • Others are wary of “yet another rewrite” and fear regressions and partial functionality.

Nouveau vs new driver

  • Concern that a new, limited‑scope driver could accelerate bitrot in Nouveau, which supports very old Nvidia hardware.
  • Some argue that supporting 1990s/early‑2000s GPUs in the new driver is unnecessary; Nouveau can remain for those.
  • Legal/firmware issues (signed firmware, reclocking) make high‑performance support for many older Nvidia cards via Nouveau effectively impossible.

Nvidia firmware, open modules, and architecture

  • Discussion notes Nvidia’s open GPU kernel modules and newer firmware (GSP) that can be redistributed and abstract more functionality.
  • Several comments suggest Red Hat’s interest is likely about:
    • Getting an in‑tree, maintainable, upstream‑style driver.
    • Avoiding tight coupling to Nvidia’s proprietary stack and firmware versions.
    • Potentially sidestepping telemetry and licensing constraints, especially in data‑center/LLM settings.
  • Some find it odd to have a third‑party driver when Nvidia ships an open‑source kernel component; others say Nvidia’s code doesn’t fit kernel norms or deployment needs.

Relationship to NVK and Mesa

  • Clarified that:
    • The new Rust driver is a kernel/display/memory driver, effectively a Nouveau replacement.
    • NVK is a Vulkan userspace driver in Mesa that generates GPU commands and uses the kernel driver underneath.
  • Both layers are needed for a complete open‑source Nvidia stack.

Rust vs C and kernel development

  • Debate over whether Rust meaningfully improves safety and maintainability for GPU drivers:
    • Pro‑Rust side: memory safety, better type systems, catching logic errors, attracting newer developers, aligning with kernel’s long‑term need for contributors.
    • Skeptical side: kernel complexity is about hardware/architecture, not language; Rust is complex, no_std is specialized; C is still widely taught and used.
  • Some argue “rewrite everything in Rust” is overhyped; others see Rust in kernel as strategically important.

Telemetry and privacy concerns

  • Some commenters suspect a driver independent of Nvidia could:
    • Eliminate or reduce mandatory telemetry and licensing checks, especially in enterprise and cloud GPU use.
  • Counterpoint: enterprises that care can already mitigate telemetry, and architectural/deployment issues are more likely the main driver.
  • Broader side discussion on:
    • Why OSS users distrust telemetry (marketing use, cross‑correlation, de‑anonymization).
    • Examples where limited, transparent telemetry (e.g., some distros/browsers) is seen as acceptable.

Linux desktop and Nvidia user experience

  • Mixed reports on proprietary driver quality:
    • Some say it “works fine” on X11 and for AI workloads.
    • Others report serious issues on Wayland: broken Vulkan/OpenGL apps, multi‑monitor problems, and mode‑setting conflicts with other GPUs (e.g., IPMI/Aspeed).
  • Nvidia’s historically weak Wayland integration (GBM vs EGLStreams) is heavily criticized; AMD is described as “just working” with Wayland.
  • Some argue poor Linux desktop support is a direct consequence of Nvidia’s business model and focus on AI/data‑center GPUs without display outputs.

Compilation time and hardware

  • Brief debate on Rust’s compile‑time cost in kernel builds:
    • Some say kernel build time on modern CPUs dwarfs any Rust overhead and most users use precompiled kernels.
    • Others note many still compile kernels (e.g., for work, experimentation, Gentoo) on older hardware and will feel the impact.
  • Suggested mitigations: cross‑compilation and ongoing effort to speed up the Rust compiler.

Broader business and ecosystem context

  • Nvidia’s dominance in AI/LLMs is seen as making Linux support critical, but mainly for servers, not desktops.
  • Nvidia is perceived as treating its driver stack and SDK as a key competitive moat; tight control over software and licensing is described as central to its model.
  • Some commenters conclude: if you care deeply about Linux desktop and open drivers, avoiding Nvidia hardware remains the safest choice.