AMD Reportedly Plans To Bring Mantle To Linux, Calls Mantle An Open-Source API

[Yorgos]

Well there's no gpu vendor with a reliable and fast Opencl implementation out there, just remember that AMD opencl can't compile large kernels without loosing all the performance yet.

support on nVidia's gpus is still at 1.1 from 2010 and nVidia has even problems with the latest DX versions, they do not have h/w support for the 11.1 and 11.2 versions, they support it via their driver.

on the other hand, what do you mean by fast implementation, fast compared to what? what is fast for you since the only one who cares about openCL is AMD, now Altera is supporting 1.0 and Arm cpus are heading toward this way.

[Nille]

As others have said about open sourcing and API's.
FYI: An API is a set of agreements about what and how something is done, it is a form of a specification not implementation.
It is not a piece of software but a specification. Meaning an API can be open but not open source.
An API can have open source implementations that are pieces of software.

Mantle has a Hardware dependent and Hardware independent part. Everything thats not Hardware dependent has to be open. And there is still a really huge problem. There is no Shader Compiler, AMD currently use the HLSL Compiler from the DX SDK.

[plonoma]

As others have said about open sourcing and API's.
FYI: An API is a set of agreements about what and how something is done, it is a form of a specification not implementation.
It is not a piece of software but a specification. Meaning an API can be open but not open source.
An API can have open source implementations that are pieces of software.

[DMJC]

Current status of OpenGL 4 in Mesa (this is why I have faith that it will be coming next year around mid year):
GL 4.0:
17/26 Completed

GL 4.1:
4/7 Completed

GL 4.2:
9/12 Completed

GL 4.3:
12/23 Completed

GL 4.4:
4/10 Completed

Total:
46/88 Completed - OpenGL 4.4 is already 40% implemented, overall OpenGL 4.0 - 4.4 is over 50% implemented.

The other thing I'll note is this: The nouveau status matrix page, if you bother to follow it, they are getting each new card generation up to running 3d code a lot faster than they first were. It's now taking weeks not months to get each new card generation supported. They are reusing a ton of code.

[Kraut]

"Gaming or nongaming should not make a difference. You want a performant, low latency, easy to mutlithread, good to debug API in all cases."
yet in some cases you need precision above all while you probably wont need it for many gaming cases. For scientific physics simulations for instance
While gaming will prefer fluent performance with low latency

There are already solutions to that like ARB_shader_precision (Core in 4.1)
Or in case of GLES there are precision specifiers that relax precision to increase performance.

This is a simply Boolean switch and not a reason to justify the stagnation of OpenGL over all this years. (Remember Longs Peak?)

[chithanh]

AMD to be believed, when their hardware is fully compatible with 4.x OpenGL AND is good choice based on quality/performance/value, so far all of those just fail when it comes to Linux. AMD has only one advantage so far. It at least has OSS drivers, unlike Nvidia.

So far, Nvidia is flagship of quality/performance/value on Linux..

OpenGL 4.0 (and some other things) are certainly nice to have, but I am very happy with how my radeon cards work today with AMD OSS drivers.

On the other hand, NVidia uses Linux to their advantage but gives not much back.

NVIDIA has an open source driver, and they're actively contributing to it. Yes nouveau is new, but it already has the 3d code implemented, it's just power management that's lagging, and NVIDIA themselves are contributing documentation to that effort. I see no reason for AMD to enjoy a special position..

Linus Torvalds himself famously raised his middle finger on NVidia for being the single worst company he has ever dealt with.

http://www.youtube.com/watch?v=iYWzMvlj2RQ

Nouveau is not a new driver. It is almost completely reverse-engineered, including the power management. NVidia contributions to nouveau are small, and not very frequent, especially compared with AMD or Intel. NVidia only release documentation as they please, and sometimes refuse to even discuss releasing documentation that was requested from them. Then you have gold nuggets like that one:

But longer-term, I think we'd like to move things to the firmware that either we're not comfortable documenting, or things that are Real Hard to get right, like P-state switching.

http://lists.freedesktop.org/archive...er/014495.html
So instead of documenting or open sourcing certain things, they will rather put the code inside GPU firmware blobs.

Untill AMD is competitive, or their technology give great advantage - I don't think their technologies will be accepted.

AMD Mantle on Windows is already by BF4 and Thief and they said that it will increase to 50 by the time that Microsoft releases DirectX 12. Many of the game engines are cross-platform, so I see no reason why they should drop Mantle when the game is ported to Linux.

[Sdar]

sure, they don't have proper performance in openCL but there is a chance to support a rival's product and do it better.

Well there's no gpu vendor with a reliable and fast Opencl implementation out there, just remember that AMD opencl can't compile large kernels without loosing all the performance yet.

[DMJC]

AMD to be believed, when their hardware is fully compatible with 4.x OpenGL AND is good choice based on quality/performance/value, so far all of those just fail when it comes to Linux. AMD has only one advantage so far. It at least has OSS drivers, unlike Nvidia.
So far, Nvidia is flagship of quality/performance/value on Linux. Untill AMD is competitive, or their technology give great advantage - I don't think their technologies will be accepted.

I agree here except for one thing. NVIDIA has an open source driver, and they're actively contributing to it. Yes nouveau is new, but it already has the 3d code implemented, it's just power management that's lagging, and NVIDIA themselves are contributing documentation to that effort. I see no reason for AMD to enjoy a special position. They abandoned Unix in the 1990s and 2000s and now that the market is swinging back they suddenly got caught with their pants down? That's their problem not ours. Intel's GL implementation works, sure it's not lightning fast but they're not trying to make high end cards.

[Figueiredo]

Then maybe they should form a new consortium between themselves, NVidia, and Intel. Possibly include game developers. I don't want an API that is controlled by one entity.

Valve seems to be the one better positioned to handle such a task. Since they are already developing a gaming OS, might as well defined the API to be used.

[UraniumDeer]

i think you jumped on hype train a bit too soon.

it is open API, not open source solution.

and you kinda require GCN to use it which is why only new amd cards are compatible.


i think only radeonsi could actually go for what you say. other vendors definitely won't license something needing complete abandonment of their way when that vendor is not a threat. amd simply has no cards to dictate tempo

I don't exactly know what I was thinking when writing about open sourcing Mantle.
Anyways, in the early days, if memory serves, Mantle only required the graphicscard to have some programmable cores, like GCN, or CUDA, so it could work with all (advanced) vendors.
I find it hard to believe that AMD would actively hinder other hardware vendors in using Mantle, as it might limit the usage of Mantle. That said, I'm quite impressed with the amount of AAA titles that is getting Mantle support, with it actually requiring newer GCN cores.