LU Decomposition on GPU MATLAB -

i trying run lu decomposition on matlab such use gpu. according nvidia/matlab documentation, lu supposed supported cuda (see, example http://www.nvidia.com/content/gtc-2010/pdfs/2267_gtc2010.pdf).

now, have compared speeds between cpu , gpu, , while gpu indeed faster matrix multiplication , fft seems give pretty same results lu decomposition, important me.

i have tried different sizes, remains pretty same.

for instance,

on gpu:

a=gpuarray(randn(1000)); tic; [l,u,p]=lu(a); toc elapsed time 0.056832 seconds. 

on cpu:

b=randn(1000); tic; [l,u,p]=lu(b); toc elapsed time 0.031463 seconds. 

cpu faster. cpu i7-2630qm , gpu gt-550m (laptop). tried on stronger computer has gtx-660 , results same.

my matlab version 2012b

using matlab r2013a on tesla c2070, see this:

a = gpuarray.randn(1000); tic; [l,u,p]=lu(a); toc elapsed time 0.016663 seconds. 

which 2x faster cpu. matrix size increases further, speedup increases, on machine peaking @ 5x faster on gpu - typical high-end (albeit old) gpu compared decent 6-core cpu.