Originally posted by mbriar
View Post
Announcement
Collapse
No announcement yet.
NVK Lands Sparse Memory Support - Enabling More Games To Run On This Open-Source Driver
Collapse
X
-
Originally posted by mbriar View Post
Yes, but to me it looks like the reason is probably that nvidia has just hardware that's much nicer to work with than intel, after all the main nvk developer also worked for intel on anv for many years prior.
- Likes 2
Comment
-
I watched one of Faith's early presentations on NVK, where she described her motivations and hopes for the project. It wasn't just to develop a Vulkan driver for NVIDIA, it was also in big part to improve Mesa's Vulkan driver infrastructure and common code.
Mesa has a lot of infrastructure around OpenGL, that has matured over many years, and it is amazing. AFAIK (from Alyssa's blogs), this is the reason why Asahi went with an OpenGL driver first. Yes, OpenGL is a more complex API, but Mesa has great frameworks for it that already take care of most of its peculiarities. So Asahi could focus on slowly reverse-engineering the hardware. Mesa didn't have anything like that for Vulkan yet.
Intel ANV was, afaik, the first Vulkan driver in Mesa. It was the pioneer that did everything from scratch. Hence, it built up a ton of legacy and bespoke weirdness over time. Other drivers like RADV started out by largely copy-pasting anything they found useful from ANV and then working from there.
Part of the purpose of NVK was to show how to make a modern FOSS Vulkan driver in Mesa, with all the experience built up over the years, and improve Mesa's frameworks to make future Vulkan driver development easier and more elegant. It did not take the same route of copy-pasting from an existing driver to get started. So it is quite different from the other Mesa drivers and really shows the state-of-the-art.
- Likes 2
Comment
Comment