AMD Isn't Done Yet Optimizing The Mesa RadeonSI Driver For Workstation OpenGL

[phoronix]

Phoronix: AMD Isn't Done Yet Optimizing The Mesa RadeonSI Driver For Workstation OpenGL

While over the past several years AMD landed numerous significant improvements to their RadeonSI driver for benefiting OpenGL workstation use-cases, that quest isn't yet over and more optimizations continue to be pursued. There are additional optimizations on the horizon for the RadeonSI Gallium3D driver for benefiting OpenGL on Linux workstations...

https://www.phoronix.com/news/RadeonSI-More-SPEC-GL-2023

[Mitch]

It's great to see AMD's work in this area.

Also, just noticed the new pic of Michael. Love it!

[edxposed]

Should replace RadeonSI with Zink and shift manpower to Zink optimizations

[coder]

Should replace RadeonSI with Zink and shift manpower to Zink optimizations

As a greedy user, I'd rather have two options: native driver + Zink, in case one or the other has a bug. Or, maybe I'm not sure if it's a bug and want to try the other for a "second opinion".

If AMD is trying to dump their proprietary userspace, then I would guess they might be reluctant to jump on the Zink bandwagon, because they have far less control over Zink. Selling workstation graphics cards that are certified for professional applications means you need a good deal of control, so that bugs, behavior changes, and performance regressions aren't introduced frequently, and you can quickly fix them if they are.

I think Zink is going to be a moving target, for a while, and changing anything in it is going to be a lot more of a negotiation, as it has to support a lot more hardware. In contrast, AMD's userspace is exclusive to AMD's hardware, and nobody is going to argue against optimizing it for that.

[cloud strife]

It's said that AMD has rewritten their proprietary OGL driver. On windows, the performance get huge improvement(https://arstechnica.com/gadgets/2022...some-pro-apps/), AMD proprietary OGL drivers are always sharing the same code base for different platforms, I am curious about the latest Linux proprietary OGL driver's performance...

[smitty3268]

Should replace RadeonSI with Zink and shift manpower to Zink optimizations

Why? How would that help AMD?

[cloud strife]

It's said that AMD has rewritten their proprietary OGL driver. On windows, the performance get huge improvement(google "Rewritten OpenGL drivers make AMD’s GPUs “up to 72%”"), AMD proprietary OGL drivers are always sharing the same code base for different platforms, I am curious about the latest Linux proprietary OGL driver's performance...

[jsnjkl]

>>" I've shown that last year in Open-Source AMD Radeon Linux Graphics In Great Shape For Workstations, Handily Beating Proprietary Driver"
It's probably not that proper that it's generating conlusion based on a very old report posted in last year. Moreover, 21.50 denotes that probably even not right proprietary workstation driver has been taken in benchmarking. AFAIK, AMD proprietary workstation driver will be labeled as "2x.Qx" like "23.Q3".

>>"That's no longer the case and their proprietary OpenGL driver on Linux has been treated as legacy for some time."
I've heard of AMD did great improvement in their AMD proprietary workstation driver on Linux recently(shall be from 23.Q1 and later?). I'm not sure how this point has been generated based on the fact that AMD proprietary workstation driver perfroms pretty competitive in both quality and performance.

[OneTimeShot]

I’m pretty sure AMD’s OpenSource and Propriety drivers are now the same code…. They have no reason to have two separate codebases

[Alexmitter]

Should replace RadeonSI with Zink and shift manpower to Zink optimizations

Why exactly would you want a worse opengl driver? Zink will frankly never beat a native opengl driver, it may get close but it will never match or beat it. Its fine as a compatibility layer but for everything else Zink is unsuited.