OpenGL state on NVIDIA/AMD

Is the state of workstation drivers as sorry?

 

What theory crafting?


In [thread=64351]"the very first post of the Mantle thread[/thread] quoting the Sept 26, 2013 presentation from DICE it says :

http://www.slideshare.net/fullscreen/DICEStudio/battlefield-4-frostbite-3-mantle/17

and also the November 2013 [post=1841332]AMD Developer Summit presentation[/post] quoted in the same thread:

http://www.slideshare.net/DICEStudio/mantle-for-developers/32


So AMD apparently have been considering Linux and MacOS support right from the start (but not game consoles, contrary to a widespread interpretation) and they have been actively pushed by ISVs to move in this direction.

AMD is still considering a Linux port as of this April, but the actual work hasn't started as they are putting their effort into making Mantle stable on Windows and improving their Linux Catalyst drivers.

http://www.phoronix.com/scan.php?page=news_item&px=MTY0MDM
https://twitter.com/NThibieroz/statuses/455736559333945344

Stating the obvious, its game developers and publishers (like Valve's Steam and SteamOS) who will have to create a gaming market, not game engine developers or tools providers - however the former cannot be successful without support from the latter.

And if Mantle is ever going to Linux, I'd actually expect any uptake to happen specifically from SteamOS/Mantle combination rather than "Mantle on Linux" in general.

 

In my opinion hell will freeze over before NVIDIA would support Mantle. They are much too arrogant for that.

 

This company can't even get things right, what makes anyone think their dead on arrival proprietary Glide 2013 API is any better?

 

Arrogance or not, it wouldn't be very wise of them to adopt an API (thereby making it a de facto standard) controlled by their main competitor. AMD was likewise well inspired to stay away from PhysX.

NVIDIA can and will support DX12. OpenGL is very likely to follow suit in some fashion.

 

Could anybody share his experience with Workstation drivers? Ethatron brought that up earlier.

Multiple developers' accounts, some public, but there's a topic specific for Mantle.

 

s/arrogant/smart/ ?

Supporting the proprietary standard of a competitor who has a head start of a couple of years is intensely stupid. Especially if there's already a vendor neutral alternative standard in the work.

Answer this question: will it take longer for Nvidia to have full complaint DX12 drivers or Mantle drivers?

Bonus question: who, do you think, will have the most rigorous compliancy tests?

There is no reason whatsoever to think that a mythical open Mantle will be any less troublesome in the compatibility department than OpenGL.

 

Drivers are tested and approved by the software developpers, i dont think an "user " of a CAD software will get much problem with any vendors in this sense.. You dont want the softwares developpers lists your brand or hardware as non wanted for their softwares...
But thoses softwares are not games, and i start to think the problem between OpenGL and drivers teams is particulary to the games aspect of it. ( In addition, they dont use all forcibly OpenGL, it depend really of the softwares )

On professionnal softwares, last "drivers" reviews was tend to show a paint totally different of what is described in this blog.

 

Well, CAD software is also written by someone, and CAD viewports are also demanding just as games. If they ask for standard-compliant behaviour, or an extension to work as documented, do they experience the same?
Or are they part of the "problem" because they insist that driver do not behave as documented but rather the same as when they had implemented a fix for a non-functional extension back in the day? Is there a hen or egg thing paralyzing (r)evolution?

 

This is the point. If Nvidia blocks any effort within Khronos Group at a clean break for OpenGL in order to maintain their vendor lock-in, the only way to overcome this is to make an alternative API and ensure its wide adoption.

If Mantle would become a de-facto high performance graphics API on Linux and Mac, other vendors will have to follow suit. But Mantle has to prove its viability on Windows first.

There are official OpenGL 4.x conformance tests? Didn't know that.

Firaxis, who are the makers of Civilization: Beyond Earth, assume otherwise.

http://programmableweb.com/news/2k-...-mantle-graphics-api/2014/04/24#ixzz326JTnD5X

"Q. The AMD Mantle website states that "Mantle makes game development easier." If this is so, how does Mantle and the Mantle API make game development easier for developers?

A. The Mantle API is able to be backed with a much simpler, faster driver. The increased efficiency means that we no longer need to add as much complexity to our rendering pipeline to achieve the required level of performance. Mantle’s simplicity also means that it will probably be more stable over the long run. We anticipate that the bug rate for Mantle to be lower than previous systems we worked with."​

The original quote above is a bit on the emotional side - if you actually look at the results of an OpenGL compatibility testing project quoted [post=1847012]in the post above[/post], OpenGL 4.0-4.2 is quite fine on recent AMD hardware, and even though there were some regressions for the OpenGL 3.x path in 2013-2014 driver releases, AMD are slowly fixing them.
http://www.g-truc.net/project-0033.html#menu


However, they currently face another significant undertaking on Linux. AMD proposed a new graphics architecture, where both Mesa/Gallium3D and their own proprietary Catalyst will work through the existing open-source RadeonDRM (Direct Rendering Manager) kernel mode driver. This however requires some significant changes to the kernel-mode driver API, and there is no guarantee that these changes will be approved by Linux kernel guys.
http://www.phoronix.com/vr.php?view=20050


So before they can start coding Mantle support (if they will), they will have to redesign their current Catalyst driver first, as it looks like they have too few Linux engineering resources to work on this simultaneously.

 

Neither did I!

 

They were only finally brought back with OpenGL 4.4 last year. Though I admittedly don't know if they are any good.

 

Mantle as it is today is not a portable API. Despite what you may read from enthusiastic devs, it needs changes to support other hardware architectures and someone needs to organize/arbitrate that process. The democratic solution (Khronos) hasn't yet proven that it can do this here. Thus how you think that any of your visions are possible without a platform vendor involved (i.e. like on windows) is beyond me.

Don't get me wrong, something Mantle-like is probably not a bad place to start iterating, but any portable API would by necessity not be what Mantle is today.

 

It's to be expected that a simpler API will have its own teething problems in its first deployment. That doesn't invalidate the fact that it may be easier to maintain and/or develop for in the long run.

But all of that is besides the point. Even if Mantle were 100% bug free from the get go, it wouldn't change a thing for other vendors. There is no justifiable reason to implement an API that will give your competitor a long lasting edge because of implementation lead time, ownership of the specification, ownership of the compliance test suite (if one exists at all) etc.

A CEO who decides to hand over such a strategic advantage to a competitor without being forced to (and, let's be honest, Mantle is a far from being there) should be fired on the spot.

 

For the Windows platform, that Mantle-like API will be obviously DX12. For Linux and Mac... no clue.

 

I think everyone understands the current state of affairs, i.e. that Mantle is not even public as of now, that there is a long road to support cross-platform development, and an even longer road to hardware portability and ultimately to an open API. And that's not even certain that AMD wants to go this road.

However it takes a first step in order to walk the walk, and there is a clear interest from the application programmers.

Remember that It took Microsoft 5 years to come up with a decent version of Direct3D, that is Direct3D 7, and developers mostly ignored it until introduction of Direct3D 5 and DrawPrimitive API.

 

No one is disagreeing that we need lower-level APIs (at least in the AAA games space). The only thing we disagree on is that you think there's some path by which AMD can just morph Mantle into a portable API across different hardware and platforms without other IHVs or platform holders being involved. I consider that an absurd notion for the reasons that I've stated.

So if you agree that platform vendors and other IHVs *do* need to be involved then we're back to what Khronos is; the question is does that model even work for this sort of breaking change? And if it doesn't, what model does work? The other one that has been proven successful on this front is Microsoft, which is a platform holder. So either you believe in the Khronos democracy model (i.e. they're just being slow or something) or you pretty much have to agree that a platform vendor has to get involved/onboard as far as I'm concerned. Everything else is just meaningless speculation.

 

Interesting thread. Here is a link from another tangent as to the state of OpenGL on various vendors.

https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/

Good read I thought.

 

Direct3D 8 sucked : no support for the Voodoo 2, LOL.

Regarding OpenGL games : that long ago, there was a subset of OpenGL that was made without asking anyone, i.e. having enough OpenGL to run Quake and Half-Life 1. That lead to drivers that could run Quake and failed at running everything else - not necessarily a good thing, but the meaning of life was to run Quake 2, Half-Life and then Counter Strike.

So, could about the same be done?, this time with OpenGL 4.4 or whatever the next version is, not OpenGL 1.1
Have some major title come out (say, Half-Life 3) and let the vendors scramble to support all the features it strictly needs.

As an aside : I have no idea what does Radeon 5000/6000 tech supports (4.0? 4.3? 4.4? 3.3?) and it would be a bummer if it doesn't get further support (for the buddy who got an APU. I should have made him get an Intel + geforce 610)