On Mon, Jan 10, 2011 at 11:35 AM, Mark Wiebe <mwwi...@gmail.com> wrote:
> I'm a bit curious why the jump from 1 to 2 threads is scaling so poorly. > Your timings have improvement factors of 1.85, 1.68, 1.64, and 1.79. Since > the computation is trivial data parallelism, and I believe it's still pretty > far off the memory bandwidth limit, I would expect a speedup of 1.95 or > higher. It looks like it is the memory bandwidth which is limiting the scalability. The slower operations scale much better than faster ones. Below are some timings of successively faster operations. When the operation is slow enough, it scales like I was expecting... -Mark Computing: 'cos(x**1.1) + sin(x**1.3) + tan(x**2.3)' with 20000000 points Using numpy: *** Time elapsed: 14.47 Using numexpr: *** Time elapsed for 1 threads: 12.659000 *** Time elapsed for 2 threads: 6.357000 *** Ratio from 1 to 2 threads: 1.991348 Using numexpr_iter: *** Time elapsed for 1 threads: 12.573000 *** Time elapsed for 2 threads: 6.398000 *** Ratio from 1 to 2 threads: 1.965145 Computing: 'x**2.345' with 20000000 points Using numpy: *** Time elapsed: 3.506 Using numexpr: *** Time elapsed for 1 threads: 3.375000 *** Time elapsed for 2 threads: 1.747000 *** Ratio from 1 to 2 threads: 1.931883 Using numexpr_iter: *** Time elapsed for 1 threads: 3.266000 *** Time elapsed for 2 threads: 1.760000 *** Ratio from 1 to 2 threads: 1.855682 Computing: '1*x+2*x+3*x+4*x+5*x+6*x+7*x+8*x+9*x+10*x+11*x+12*x+13*x+14*x' with 20000000 points Using numpy: *** Time elapsed: 9.774 Using numexpr: *** Time elapsed for 1 threads: 1.314000 *** Time elapsed for 2 threads: 0.703000 *** Ratio from 1 to 2 threads: 1.869132 Using numexpr_iter: *** Time elapsed for 1 threads: 1.257000 *** Time elapsed for 2 threads: 0.683000 *** Ratio from 1 to 2 threads: 1.840410 Computing: 'x+2.345' with 20000000 points Using numpy: *** Time elapsed: 0.343 Using numexpr: *** Time elapsed for 1 threads: 0.348000 *** Time elapsed for 2 threads: 0.300000 *** Ratio from 1 to 2 threads: 1.160000 Using numexpr_iter: *** Time elapsed for 1 threads: 0.354000 *** Time elapsed for 2 threads: 0.293000 *** Ratio from 1 to 2 threads: 1.208191
_______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
