AMD Is Still Moving Towards A Unified Open-Source Driver

[schmidtbag]

But what about multiple GPUs with different OpenGL version support and OpenGL extension support? You'd always be restricted to the lowest common features...

That's why I said "assuming all GPUs are compatible with the rendering, and assuming a hardware bridge isn't required". I'm not saying mix and matching GPUs and brands is recommended, but if you know what you're doing it'd be a great way to get extra performance without abandoning any hardware.

[haagch]

But what about multiple GPUs with different OpenGL version support and OpenGL extension support? You'd always be restricted to the lowest common features...

[schmidtbag]

I recall a Mesa developer mentioning that crossfire support has already been possible with open source drivers for a long time, just that nobody has stepped up to do the hard work (coding). But AFAIK there are no proprietary / unknown bits that inherently prevent an open source implementation.

Well sure, of course crossfire was possible for a long time. Crossfire is actually quite simple - with AFR, you're just rendering frames on alternating GPUs. Theoretically you could use any GPU combination (assuming all GPUs are compatible with the rendering, and assuming a hardware bridge isn't required). If you were to mix and match GPUs you'd just get very undesirable results, which is probably why you currently can't do that. But if there becomes a Mesa extension allowing you to do AFR regardless of brand, linux will gain some HUGE interest in the gaming industry. Think about it - you could do multi-GPU setups involving intel graphics and put those IGPs to good use without involving something like Enduro/Optimus.

[Ancurio]

I seriously hope this means crossfire support. I just installed catalyst for the first time in about year yesterday and crossfire currently only works on the Heaven benchmark, and even at that it performs horribly. Crossfire is the only reason I switched to catalyst in the first place, and the lack of crossfire is the one thing preventing me from playing some games at 60+FPS at high(er) detail settings.

I recall a Mesa developer mentioning that crossfire support has already been possible with open source drivers for a long time, just that nobody has stepped up to do the hard work (coding). But AFAIK there are no proprietary / unknown bits that inherently prevent an open source implementation.

[schmidtbag]

This might bring CrossFire to Mesa... Imagine "CrossFire" with totally different GPUs using Mesa, it would totally rock.

I seriously hope this means crossfire support. I just installed catalyst for the first time in about year yesterday and crossfire currently only works on the Heaven benchmark, and even at that it performs horribly. Crossfire is the only reason I switched to catalyst in the first place, and the lack of crossfire is the one thing preventing me from playing some games at 60+FPS at high(er) detail settings.

[System25]

Nice idea, but...

Nice idea, but ideally I can wish them not to stop half way and consider opensource driver as primary target at Linux, just as Intel did. But still, nice idea since it is kernel module what is attrubuted to most Linux troubles.

[verde]

That IS a solution for AMD GPUs. Lets se...

[przemoli]

Reason for wanting OpenGL 4.4 is AZDO.

But here is the catch... If You do not need tesselation? Mesa will have AZDO faster then OpenGL 4.4

After all AZDO is "just" a bunch of extensions, and Mesa is famous for cherry picking which ext. gets support next

So OpenGL 3.3 + few extra extensions should do the trick (Or 4.0 + few extensions if tessel is necessary!)


"I MUSTA HAVEA 4.4" is relict of monolithic proprietary drivers, where company politics dictate gradual development (no 4.4 before 4.3 is finished, no 4.3 before 4.2 is finished, and oh that hw wont get 3.2 .. because!)
And "I MUST HAVEA Compatibility Profile" is another sad example.

(Thankfully, Mesa allow for overriding both supported version and supported profile... And then some software just magically start to work :P )

[OneTimeShot]

Actually all hardware that supports OpenGL 3.3 supports OpenGL up to 4.4 with proprietary drivers from AMD, Nvidia and Intel(on Windows for now). Something that doesn't support it is realy old hardware already and in 2017 not many gamers will have such and even OSS drivers will have support of it. So if someone would like to create modern and powerful engine that it should be ok. Ofcourse "powerful" engine doesn't sound nice when you have 1 developer but its his choice anyway.

Support-ish, some parts of newer OpenGL don't need hardware support, but a lot of it does. It might report OGL 4.4, but if the hardware physically isn't there, there is no magic in the proprietary drivers that will be better than software rasterization.

The debate is a little silly, though. If you spend $8M dollars developing a game, most of that will be graphics artists. You can afford to pay a dev for 6 months to produce "a lower quality graphics engine" that will work on Windows Vista era graphics cards (== OpenGL 3.3). If that is pushing your budget, realistically you'll not support Linux long before you don't support people on Windows XP...

[sunweb]

Nobody at this point would develop any game or engine which has 4.4 as the minimum version.

Actually all hardware that supports OpenGL 3.3 supports OpenGL up to 4.4 with proprietary drivers from AMD, Nvidia and Intel(on Windows for now). Something that doesn't support it is realy old hardware already and in 2017 not many gamers will have such and even OSS drivers will have support of it. So if someone would like to create modern and powerful engine that it should be ok. Ofcourse "powerful" engine doesn't sound nice when you have 1 developer but its his choice anyway.