Zink Lands Threaded Context Support For A Big Speed Boost With OpenGL Over Vulkan

[phoronix]

Phoronix: Zink Lands Threaded Context Support For A Big Speed Boost With OpenGL Over Vulkan

Mesa 21.1 is looking to be another exciting release to be introduced later this quarter while going into feature freeze around mid-April. The latest work to land is threaded context support for Zink, which means faster performance for this OpenGL-over-Vulkan implementation...

https://www.phoronix.com/scan.php?page=news_item&px=Zink-Lands-Threaded-Ctx

[emblemparade]

Smashing success.

Some of you might be thinking that 70% is not good enough. But 10 years from now we'll be able to play OpenGL games at resolutions and framerates far greater than they were originally designed for, even though none of our drivers will support this antiquated and weird API.

Mike Blumenkrantz, you are doing god's work.

[darkbasic]

70% is definitely not enough and it will never be. I believe Zink can (and will) achieve much, much better results.

[oiaohm]

70% is definitely not enough and it will never be. I believe Zink can (and will) achieve much, much better results.

You really do need to rethink that. At 60-70% Zink to run like GTK opengl accelerated rendering provided advantage over software rendering on the CPU by a large margin. There are many old games that are more the play able at 60-70% percent on modern GPUs.

Case where the program will not work right due to some vendor Opengl issue 60-70 percent performance is still way better than crash out all the time.

So for particular use cases 60-70% is good enough as in playable or allows work to be performed.

This is not that I am not wishing for more it would be good if in time Zink could get up into the 80-90% range..

darkbasic there is more than 1 type of user here. Business users not gaming desktop case like wanting opengl rendering in libreoffice and so on it work 60-70% is in the good enough range.

This is starting crossing the threshold for parties like libreoffice to be asking: Should they bother with unique code to deal with Nvidia quirks or should they just recommend users on Nvidia use Zink? This is a new form of questions.

[R41N3R]

Great work! Zink did progress quite a lot in Mesa mainline. It will be interesting to see benchmarks this time as 60% to 70% performance of a native OpenGL driver sounds good enough for many use cases :-)

[darkbasic]

darkbasic there is more than 1 type of user here. Business users not gaming desktop case like wanting opengl rendering in libreoffice and so on it work 60-70% is in the good enough range.

This is starting crossing the threshold for parties like libreoffice to be asking: Should they bother with unique code to deal with Nvidia quirks or should they just recommend users on Nvidia use Zink? This is a new form of questions.

I am talking about getting rid of OpenGL drivers altogether while you talk about circumventing vendor specific quirks in their drivers: our goals are different and thus the requirements.

[V1tol]

You forget to mention than Zink was introduced less than 2 years ago and already achieved 60-70% of performance. OpenGL implementations were polished for more than 20 years. I think in 2 years it will be 90-100% already, if not more. Right in time when no new software will use this API and it will be needed only for backwards compatibility.

For example, I remember about such proprietary thing like MoltenGL - OpenGL ES implementation on top of Metal for Apple devices. They state it is capable of doing 3x performance of native drivers. Yes I know that native OpenGL is horrible on Apple devices and Mesa is very good, but it is possible to achieve the same and do even more in Zink.=

[oleid]

You forget to mention than Zink was introduced less than 2 years ago and already achieved 60-70% of performance. OpenGL implementations were polished for more than 20 years

And you forget that Zink is re-using the already fine-tuned gallium stack.

[oiaohm]

I am talking about getting rid of OpenGL drivers altogether while you talk about circumventing vendor specific quirks in their drivers: our goals are different and thus the requirements.

Not completely. The reality is there will be thresholds. The question is how far into the future are we talking. There are historic examples like glide API that today is always emulated if it need. Yes the glide to opengl and directx and vulkan is not 100 percent efficiency and some of them are under 60% when you compare the computing power of the modern GPU vs frames given vs original Voodoo/3dfx cards.

The threshold point for not needing Opengl drivers any more will depend:
1) On how much the GPU end users have gain in performance.
2) How often GPU vendor quirks come up. If the reality is majority of users cannot run programs on vendor Opengl the fact Zink is slower comes less important.
3) And what the efficiency value of Zink is.
4) Feature completeness of Zink
5) If vendors decide to stop providing opengl drivers or not.
6) How many years into the future we are talking-how much GPU improvement first point is linked to this one..

History of glide and other emulated graphical stacks set the bar min in performance for replacement by emulation on a different graphical stack at about 50%. 60%-70% of Zink is over that threshold. Of course the worse this percentage the more generations in GPU improvements we will need to make up the difference. At 50% you would be guessing a 15-18 into the future. 10 years into the future at 60-70% is possible but more realistic if the percentage gets 80-90% for a decade into the future. 60-70% you would be talking 15 years into the future for possible replacement. Yes glide emulation with high exactness with GPU effenecy around 60-70 is seen used by end users about 15 years after. So there are examples you can look at to get some ball park figures on when.

Big issue for why Zink cannot replace drivers as a generic replacement yet is feature completeness. OpenGL ES 3.2 is still missing 3 extensions.

https://mesamatrix.net/#ExtensionsthatarenotpartofanyOpenGLorOpenGLESversion

Yes in extensions that are not part of opengl standard Zink is about 15 behind this does make a lot of applications not work.

Circumventing vendor specific/card quirks also links to how glide emulation starts as well. The early stage to driver API being replaced with emulated normally starts off with quirk elimination or cross platform support.

Sorry to say darkbasic history does not agree with the idea that 70% is not enough for direct driver support replacement with the emulated solution instead history says that 60-70% is possible enough if we are talking decade and half into the future. emblemparade decade claim looking at current values is optimistic to the extreme but if Zink keeps on improving may not be optimistic we are talking a 20% uplift here so 60-70 percent comes 80-90% and that may not be possible time will tell.

[uid313]

60-70% is cool. Is it possible for Zink to reach 100% performance?
Is it possible for Zink to exceed 100% of native OpenGL drivers?