Editing Openai/69676758-d944-8003-a3d4-835e33b4e372 (section)

=== Assistant: Yes. ===
Yes. GPU manufacturers write the primary drivers for their hardware.

More precisely and with the important nuances:

==== ### ====
* Writes and maintains: - Windows drivers - Linux proprietary drivers
* Full control over hardware documentation and behavior

===== - Writes and maintains: - Windows drivers - The official Linux driver stack (AMDGPU kernel driver + Mesa user-space components) =====
* Linux drivers are open source, but still authored and governed by AMD

===== - Writes and maintains: - Windows drivers - Linux drivers (fully open source, upstream-first) =====
* Intel GPUs are the most openly documented

==== - Microsoft, Linux, Apple do not write GPU drivers from scratch ====
* They define: - Driver models (WDDM, DRM/KMS, IOKit) - Validation, signing, and integration rules
* They provide interfaces, not hardware-specific code

==== GPU drivers are split into layers: ====

===== - Memory management =====
* Scheduling
* Security / isolation
* Display control (modesetting)

===== - Implements APIs: - Vulkan - OpenGL - DirectX =====
* Translates API calls into GPU command streams

Both layers are vendor-specific.

==== Common misconception: ====
* Mesa drivers ≠ community guesses
* Mesa contains vendor-authored or vendor-blessed code
* Even reverse-engineered drivers (e.g., Nouveau) trail far behind

==== This is exactly why: ====
* You cannot “go below Vulkan”
* Display engines, PCIe, DP, scanout are all hidden
* GPU vendors expose only what they want you to see

==== GPU vendors write their own drivers; the OS provides the framework, not the implementation. ====

If you want, I can explain how Vulkan drivers differ from DirectX drivers internally, or why driver bugs look the way they do from an engine developer’s perspective.