RadeonSI OpenGL CTS Tests Running ~30% Faster With Mesa 24.1

[aufkrawall]

Getting faster means regressing to you?

Does talking nonsense mean talking sense to you? I thought so.

[SofS]

About Zink, this discussion is not necessary. It all boils down to the limited manpower/resources available for development. It makes sense for new architectures to use it as a backend. However, if there are native paths already available or if there is enough manpower/resources then there is no reason not to use the native implementation.

This is open source we are talking about, the notion of reducing things to one solution should be abhorrent by principle. The more alternative implementations available to compete the better. Any solution pretending to be a silver bullet that completely replaces everything else doing the same thing should be avoided, it is a clear red flag.

[ResponseWriter]

As has been stated before, having multiple implementations can be beneficial for testing multiple paths to help isolate an issue. Likewise, finding potential optimisations where a particular benchmark runs faster on one than the other.

[rmfx]

I know, but that still doesn't change the fact that RadeonSi is the fastest, most optimized and mature OpenGL driver currently available. With ACO support for RadeonSi finished in Mesa 24.0, it should now be even faster. I've already noticed slight gain in 2 games with ACO.



Unless you don't own an AMD GPU, I personally find it really weird not wanting to get the most out of your GPU. In other cases like Nouveau or Asahi when there is no mature and performant OpenGL driver? Sure, Zink makes much more sense.

It prefer by far 10 vulkan drivers and only zink in mesa, rather than 10 vulkan drivers + 10 opengl drivers to maintain.
When Zink is the only GL driver left, it will get full attention and potential optimization and be way good enough to handle the aging gl applications and games.
Even if RadeonSi can squish at few percents over Radv + Zink, I prefer leaving gl to Zink still, because that means vulkan drivers (that are the future) get more attention from mesa devs.
Here is my take.

PS: I have nvidia, so yes, in my case that'll be nvk (if good enough) + zink.

[Daktyl198]

I would have thought Zink to be much more compatible and performant at this point, disappointing imho. It just works poorly compared to where DXVK got much earlier in its development. Which is a pity, as I suppose lots of users would like to get rid of the most regressing component in their system after the Linux kernel, which probably is RadeonSI...

OpenGL is a fucking mess. Most developers absolutely hate it and for good reason. There's a reason it took so long to implement OpenGL feature sets. D3D10-11 are significantly easier to work with and are much easier to implement. I'm impressed Zink has managed as much as it has, to be honest.

[Dukenukemx]

Which is a pity, as I suppose lots of users would like to get rid of the most regressing component in their system after the Linux kernel, which probably is RadeonSI...

You think RadeonSI is regressing and that's second to the Linux kernel? So you use a M2 Macbook Air or you on the M1 still? You gotta be an Apple user for sure.

[Dukenukemx]

It prefer by far 10 vulkan drivers and only zink in mesa, rather than 10 vulkan drivers + 10 opengl drivers to maintain.
When Zink is the only GL driver left, it will get full attention and potential optimization and be way good enough to handle the aging gl applications and games.
Even if RadeonSi can squish at few percents over Radv + Zink, I prefer leaving gl to Zink still, because that means vulkan drivers (that are the future) get more attention from mesa devs.
Here is my take.

As much as you want to move to the future, the OpenGL API has been around for decades and you think it's super easy to just dump it over Vulkan? As great as it sounds to make the few developers jobs easier who work on mesa, but you're asking for all the other developers to implement Vulkan when OpenGL was working fine and really doesn't need the capabilities of Vulkan? The math doesn't check out.

This is a big problem with the Linux community, because they have an ideology that they want everyone else to enforce, and what always happens is that nobody does. So of course things either don't work, or they work slowly. The people who don't use Linux will point and laugh because we can't run a game from 2005 that runs on OpenGL. We're at 4% market share, so lets not try to mess this up now.

PS: I have nvidia, so yes, in my case that'll be nvk (if good enough) + zink.

Are you using Nouveau or Nvidia's drivers? You better be practicing what you're preaching.

[rmfx]

As much as you want to move to the future, the OpenGL API has been around for decades and you think it's super easy to just dump it over Vulkan? As great as it sounds to make the few developers jobs easier who work on mesa, but you're asking for all the other developers to implement Vulkan when OpenGL was working fine and really doesn't need the capabilities of Vulkan? The math doesn't check out.

This is a big problem with the Linux community, because they have an ideology that they want everyone else to enforce, and what always happens is that nobody does. So of course things either don't work, or they work slowly. The people who don't use Linux will point and laugh because we can't run a game from 2005 that runs on OpenGL. We're at 4% market share, so lets not try to mess this up now.

Are you using Nouveau or Nvidia's drivers? You better be practicing what you're preaching.

I never said drop opengl... I said opengl support should be left to zink from now on, to allocate ressources in a much smarter and less redundant way.

I use the blob drivers for now. Are you serious? mesa just started becoming “maybe decent” regarding nvidia. Nothing is done yet. However, I use radv and zink on my old laptop. And I am happy with it.

[Dukenukemx]

I never said drop opengl... I said opengl support should be left to zink from now on, to allocate ressources in a much smarter and less redundant way.

How much hardware you think has Vulkan to make use of Zink? Even still, Zink will never 100% catch up to RadeonSi in terms of performance. Which isn't really acceptable for some people.

I use the blob drivers for now. Are you serious? mesa just started becoming “maybe decent” regarding nvidia. Nothing is done yet. However, I use radv and zink on my old laptop. And I am happy with it.

What's wrong with Nouveau? I'm sure losing a bit of performance isn't a big deal, because why waste resources on two drivers? There are reasons why we keep RadeonSi around. If you do use Zink on AMD hardware then you're just wasting system resources, because you have a better alternative. Translation layers are always worse than native. I wish someone would make a Gallium Eleven and Twelve just for this reason.

[cb88]

Translation layers are always worse than native.

That isn't true, it really depends, also Vulkan itself is lower level than even gallium... it should be possible to eliminate gallium from Mesa3d eventually and the driver acutally be faster and better, and lower maintenance. There is something to be said for different code paths but I'm not sure that is really the case since a lot of the underpinnings end up being the same for both paths.

ZINK > Vulkan
and
"Native" > Gallium have the same level of abstraction.

Acutally there are many cases where ZINK is already faster now ... and where it ends up not being faster is where there is a lack of optimisations that lead it to choking at higher FPS on say the 7900XTX, while in the same benchmarks it is faster than "native" on slower GPUs.