AMD Wants Mantle On Linux, OS X, Mobile Devices

[Marc Driftmeyer]

Note: lame 1 minute barrier.

Note: Might I add as a NeXT/Apple alum, ``Apple won't give too ****s about it'' replacing any of their OpenGL/OpenCL code that has been completely updated and modernized for OS X 10.9 and iOS 7.x.

[Marc Driftmeyer]

This won't supplant OpenGL.

If it is lucky, it will be a tool that is used in places that make sense and no where else. Most importantly, OpenGL/OpenCL were designed with interoperability in mind. AMD is all in on all 3 areas: OpenGL/OpenCL and Mantle.

Most likely they are trying to make a hybrid of OpenGL/OpenCL and calling it Mantle.

If it makes game engines more efficient, bravo! I guarantee you FEA/CFD and more applications aren't going to spin their wheels for 5% -10% performance improvements by switching to Mantle when they will just mature their OpenGL/OpenCL code.

[rvdboom]

I think Mantle could be the way to go, if it wasn't tied up to AMD (hardware).

This is how I see it :

1/ Several game engine developers have already added Mantle to their engine, and the others seem quite interested.
2/ Even though Mantle is not in XBox or PS4 in itself, it would be quite logical if AMD has designed it in such way to allow the move from these to Mantle in the easiest possible way.
2/ Three of the main game developers (DICE, Id and Epic) have stated that it would be **very bad** if there was a different Mantle-like API for each hardware provider.
3/ Mantle is supposed to be able to support other architectures with plugins.

In other words, Mantle has many things to be quickly integrated in all game engines, game developpers just want one of those low-level APIs on PC and Mantle can be extended to other architectures.
So my guess is that it won't stay tied up to AMD very long.

[log0]

I have to say I like what I've seen so far (high level bits). They are exposing GPU engines (Graphics, Compute, DMA) and memory to the developer, allow to feed the command buffers directly and handle memory allocation. This means a bit more work for developers using this API, but also more control over GPU (and simpler drivers ).

I think Mantle could be the way to go, if it wasn't tied up to AMD (hardware).

To AMD/DICE: Where is the fscking API spec?

[rvdboom]

You guys should check the provided slides above and look for instance at semiaccurate.com's forum for informed thinking about Mantle before dismissing it for various reasons.
The facts are :

1/ Game developpers and particularly engine developpers seem **very** interested in it
2/ In particular, Mantle adresses CPU overhead in OpenGL and DirectX, which is one of the main reasons Mesa is slower according to X11 developpers (IRC)
3/ There are Nvidia developpers at the AMD APU13 Summit right now
4/ The improvements provided by Mantle will apply not only to games but to any 3D application.
5/ Creating a Mantle path in a game engine seems simple (2 man months for Oxide)
6/ AMD stated that Mantle is not restricted in itself to their own architecture. It can be used with others using a plugin-system, so there's no reason NVidia and Intel cannot develop a layer to use with their architecture. And since AMD has a rather good track record in opening their standards (x86_64, Hypertransport, HSA, etc.) and passing the IP to foundation, Mantle could probably become an open standard as OpenGL.

I don't know if Mantle will live to its promises but I do think that people should look into it closely before dismissing it completely.

[Nille]

Thats their problem basically.

If you want high speed mantle rendering you can use the fglrx. A Gallium3D State Tracker change nothing. But if Intel can Support Mantle on linux is a big win.

[[Knuckles]]

I am really confused at this point. Why would anyone code against a vendor specific API? In Linux, you code against OpenGL, and it will work on pretty much any graphics card.

Oh, but you see, that's not true. In reality, you code against OpenGL, and it will work on nvidia, intel and mesa. But not on Catalyst.
Just check the release notes of every game released on linux for the past 12 months.

So instead of cleaning up their act, they just introduce another API and say "you shouldn't use OpenGL on our hardware, anyway".

[89c51]

You sure? I mean Intel doens't use Gallium3D

Thats their problem basically.

[Nille]

A G3D state tracker will be more than welcome if they are serious about linux.

You sure? I mean Intel doens't use Gallium3D

[89c51]

A G3D state tracker will be more than welcome if they are serious about linux.



And an analysis by elanthis.