PS4 Blowout: GPGPU, DirectX 11; Sony London to “Set the Bar for the Industry”, Global Illumination, Instant Radiosity, More

[zarx]

Wlakiz said:
as someone who has done CUDA programming, I can tell you that GPGPU programming no easier than using cell. Objects needs to be copied to the GPU's memory, which could range from fast but small local memory to the slow but larger shared memory. When you copy back, the memory needs to coalesced to achieve optimal write performance. Not to mention, you need to manage and sync warps (cluster of threads) which can cause huge bottlenecks if you require all data before sending stuff back to the main memory.

Of course, this is all from 3 years ago.. not sure how GPGPU programming changed these days.

Shouldn't AMD's HSA (Heterogeneous Systems Architecture) inititives and memory architecture in consoles make it a lot easier? I know AMD is aiming for full HSA support in it's 2014 GPUs in the PC space, but I imagine it would be much easier to implement in a console espicialy if as rumoured they are providing the GPU and CPUs for the next Xbox and playstation and they will both presumably have unified RAM (I don't know how eDRAM etc fits into that). 

I don't know much about it other than a few articles I have read myself.