Best GPGPU framework

[Mark R]

I'm going to need to use the GPU for a project for reasons of brute force computation.

I've only ever used DirectX, and while my work is image based, the computations are such that they would be best done with a more general framework, and while they could be done with pixel shaders, I'm looking for a more manageable solution.

As I have AMD hardware, I was thinking of going for OpenCL rather than CUDA - but I've been rather behind the times and just found out about Vulkan.

There are c# frameworks for both Vulkan and OpenCL, so that is convenient for me. Are there any meaningful differences between the two frameworks that should push me one way or the other?

 

[William Gaatjes]

I'm going to need to use the GPU for a project for reasons of brute force computation.

I've only ever used DirectX, and while my work is image based, the computations are such that they would be best done with a more general framework, and while they could be done with pixel shaders, I'm looking for a more manageable solution.

As I have AMD hardware, I was thinking of going for OpenCL rather than CUDA - but I've been rather behind the times and just found out about Vulkan.

There are c# frameworks for both Vulkan and OpenCL, so that is convenient for me. Are there any meaningful differences between the two frameworks that should push me one way or the other?

I do not know if this is what you seek :

http://www.anandtech.com/show/9792/...e-announced-c-and-cuda-compilers-for-amd-gpus

https://forum.beyond3d.com/threads/...c-compiler-w-cuda-interop-for-amd-gpus.57412/

Maybe ?

 

[veri745]

I haven't use Vulkan, but the last time I used OpenCL, there was a crap-ton of boilerplate required to get it off the ground.

Otherwise OpenCL is quite powerful and mostly straightforward. I'd use it again if I had the need.

 

[Cogman]

I haven't use Vulkan, but the last time I used OpenCL, there was a crap-ton of boilerplate required to get it off the ground.

Otherwise OpenCL is quite powerful and mostly straightforward. I'd use it again if I had the need.

Vulkan, from what I hear, is no different (But I've used neither).

That being said, it sound like they are due to be one and the same thing. So I would say for newness sake, do it in Vulkan. For stability sake, OpenCL would be the best choice.

Cuda I hear is a better API than OpenCL, but it is nVidia only.

 

[Tugrul_8192bit]

From OpenCL 1.2 to 2.0, C# GPGPU environment for multiple GPU loadbalancing(kernel partitioning) or task pooling, pipelining and whatever type of dynamic parallelism is needed:

Download: https://github.com/tugrul512bit/Cekirdekler

Wiki: https://github.com/tugrul512bit/Cekirdekler/wiki

Example: