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

[Aryma]

why directx over vulkan is faster then opengl over vulkan ?

how zink speed compare it to wined3d ?

[dragon321]

I've performed some quick test and Zink achieved about 60% of native driver performance. Considering the fact Zink used to be much slower it's really nice score. Good job, I hope more improvements will come soon.

[TemplarGR]

Even 50% of native performance would be enough. People forget that every year gpus gain some performance on average. And that new AAA games that push the envelope in terms of performance required are less likely to be OpenGL in the future, since all are Vulkan now. So OpenGL will be mostly useful for playing existing games and/or less demanding indies. So the difference in performance compared to native wouldn't be so noticeable in the future. And i am not talking a decade, even 5-6 years are enough.

[bridgman]

From Mike's blog post:

Special thanks to Vulkan crash test dummy and uninitialized variable enthusiast Lionel Landwerlin for tripping and falling over basically every line of code in this implementation to help get it to the finish line for your consumption.

That is the best description of testing alpha-level software I have seen in a long time

[oiaohm]

Even 50% of native performance would be enough. People forget that every year gpus gain some performance on average. And that new AAA games that push the envelope in terms of performance required are less likely to be OpenGL in the future, since all are Vulkan now. So OpenGL will be mostly useful for playing existing games and/or less demanding indies. So the difference in performance compared to native wouldn't be so noticeable in the future. And i am not talking a decade, even 5-6 years are enough.

https://www.dsogaming.com/news/nvidias-performance-improvement-per-generation-has-dropped-from-60-pre-2010-to-30-post-2010/

The reality is the high end GPU slowed their performance increase in 2010. The gains year to year are reducing. When you get into the lower end GPU it will take a decade to a decade and a half to get 100% performance increase to cover 50% of native performance.

If were not in a performance gain slowdown 5-6 may be right. With price of GPU look like it will go up more people will be stuck in the lower end GPU with no performance increase to prior generations. Yes the covid-19 mess and the after effects on the GPU market combined with the slow down in gain per generation of GPU make the 5-6 years bit highly optimistic with 50% performance will be use-able to general users.

[dokipen]

Another benefit we might see from zink is that we might be able to get extensions that are not core (which should have been IMHO) into zink even on gpus from other vendors that they weren't designed for. Imagine NV_Command_List on intel, amd and even apple M1 gpus. Also imagine bindless textures everywhere! Just having a conformant, stable, predictable api will be a big plus for me. OpenGL still has a large part to play in computer graphics education for a while I think. Noone starting out wan't to mess around with vulkan's boilerplate. It just sucks the fun out of it.

This % of performance discussion doesn't really make sense at this point. We need to give the project time to get out there, settle and find it's place. I myself am looking forward to graphics programming on mac as I use a mac mini as my daily driver. I compiled a list of things that moltenvk needs to support for zink to work on mac as there are a few things missing still....

https://docs.google.com/spreadsheets/d/18kruiFWy9EOItQGYeWMk4B1c5cH_5JM61GSaWe0DKHU/edit?usp=sharing

Had a chat with mike and eric on twitter and the main feature that will be hard is geometry shaders as metal doesn't have them. They will have to be emulated using shaders which is a big job. The other issues are apparently easier to emulate and I think they are hoping moltenvk will provide a lot of the needed features.

Exciting times and impressive work.

[TemplarGR]

https://www.dsogaming.com/news/nvidias-performance-improvement-per-generation-has-dropped-from-60-pre-2010-to-30-post-2010/

The reality is the high end GPU slowed their performance increase in 2010. The gains year to year are reducing. When you get into the lower end GPU it will take a decade to a decade and a half to get 100% performance increase to cover 50% of native performance.

If were not in a performance gain slowdown 5-6 may be right. With price of GPU look like it will go up more people will be stuck in the lower end GPU with no performance increase to prior generations. Yes the covid-19 mess and the after effects on the GPU market combined with the slow down in gain per generation of GPU make the 5-6 years bit highly optimistic with 50% performance will be use-able to general users.

GPU prices will become cheap as dirt once more competition enters the market (Intel and/or bigger ARM competitors) and the bitcoin ponzi scheme collapses eventually.

[oiaohm]

GPU prices will become cheap as dirt once more competition enters the market (Intel and/or bigger ARM competitors) and the bitcoin ponzi scheme collapses eventually.

That is a no we hoped Intel was going to use their own fabs for their GPU products but that is no they are not. TSMC will be fab for the Intel and most of the ARM GPU as well. The current problem that sees as screwed for about 5 years there is not enough fab space. Even without the crypto currency mining boom.

10 years is where we are in the current mess. Its going to take 2 to 3 years to bring on new fabs then about another 5 to catch up on the backlog. TSMC has cease cost reduction on older nodes at the moment to cover the cost of building the new fabs.

The horrible reality is the majority of AMD/Nvidia and future Intel GPU will come from one company TSMC and that is for the next 5 years at least. Cost of silicon from that TSMC fabs is likely to go up not down in the short term.

Yes GPU prices coming cheap as dirt unless something changes that not for at least another 7-8 years.

Covid-19 has increased the number of computers need in the home as well as the mining boom. Covid-19 has increased business need for servers as well.

Like it or not the silicon market at the moment is nicely screwed.

[haagch]

I recently came about an unexpected benefit of zink when I started an OpenGL app in xephyr, which traditionally does not support glx hardware acceleration.

https://i.imgur.com/3MVCkaW.jpg

[bridgman]

GPU prices will become cheap as dirt once more competition enters the market (Intel and/or bigger ARM competitors) and the bitcoin ponzi scheme collapses eventually.

That might be optimistic. We are no longer seeing the kind of drops in cost per transistor that drove most of the historical price/performance gains.

That said, the crazy retail/scalper pricing we are seeing today should go away with the next mining bust for sure.