Page 1 of 1

AMD Mantle

Posted: Wed Dec 18, 2013 9:50 pm
by Valentin N
Hey,

I wonder if you will implement AMD's Mantle API in future Core Engine?

Re: AMD Mantle

Posted: Wed Dec 18, 2013 10:18 pm
by mdk777
not likely as open cl already serves the function.
Mantle augments or replaces MS Direct X functions that are a bottle neck.
These functions are not used for FOLDING. :mrgreen:

Re: AMD Mantle

Posted: Thu Dec 19, 2013 10:16 am
by powerarmour
Mantle is a low level 'graphics' API, OpenCL is already a low level 'compute' API

Re: AMD Mantle

Posted: Thu Dec 19, 2013 1:12 pm
by Valentin N
I am aware of that but mantle is optimized for GCN-architecture and wouldn't AMD folder benefit from that?

Re: AMD Mantle

Posted: Thu Dec 19, 2013 3:01 pm
by mdk777
http://developer.amd.com/tools-and-sdks ... computing/

Here are the programs, compilers, and math libraries along with the sdk that programers can use to for compute.

In theory, they are optimized for for GCN-architecture.

Of course, theory and practice are not always the same thing. It has been around a year and the AMD opencl functions still don't work with the popular rendering program BLENDER. :?

Mantle is a program to optimize the communication between the cpu and gpu, to co-ordinate graphic instructions(game scene/display information) .

could it help in some way co-ordinate instructions to preform open cl?

perhaps, but it would only be as an aside, a possible lucky off label application as it were. :mrgreen:

Re: AMD Mantle

Posted: Thu Dec 19, 2013 3:08 pm
by bruce
Software that is designed to create graphics images can be rewritten to use Mantle rather than DirectX or OpenGL. That software will probably benefit from the change. Somebody might rewrite FAH's Viewer to run more effectively, especially if Mantle runs on OS-X, on Linux, and on Windows. Mantle will not benefit the FahCore software because it doesn't generate graphics images, it does scientific computations using OpenCL.

Will AMD continue to support OpenCL or will they create a proprietary Compute language like NV's CUDA? (I have not seen any suggestions for that change in direction) ... or will OpenCL be optimized for GCN? (I don't know) ... and in either case, will that new set of drivers run less efficiently on older GPUs? (Again, I don't know.)