Multi-GPU on MATLAB 2013

Salam, I am trying to do a MultiGPU implementation in MATLAB2013. But when I copy my array (say A) which i need to divide on multiple gpus, the copy is only made to one gpu, not all. Then I tried to set 4 gpus to 4 matlab workers separately, but the gpuDevice(i) is called inside the parfor loop, and every time the iteration starts, the gpu initializes, and it takes more time than single gpu to implement my code. Kindly help

Also i Attached a different m-file setGPU

function setGPU()

n = (matlabpool('size'));

parfor idx=1:n
     t = getCurrentTask()
 t.ID
if(t.ID<3)
    gpuDevice(1)
end
if(t.ID>2 && t.ID<5)
    gpuDevice(2)
end
if(t.ID>4 && t.ID<7)
    gpuDevice(3)
end
if(t.ID>6 && t.ID<=8)
    gpuDevice(4)
end
end

and called it in my matlabpool function like this:

matlabpool ('open' ,2,'AttachedFiles',setGPU)

but still i can't set 4 separate gpus to 4 separate workers.. any solution ?

Answer 1

[Edric Ellis]

Why not use SPMD to do this once after you've opened the pool?

matlabpool(...);
spmd
  n = numlabs;
  if labindex < 3
    gpuDevice(1);
  end
  if labindex > 2 && labindex < 5
    gpuDevice(2);
  end
  ...
end