Roll the bike forward in as straight a line as possible while sitting on the saddle, measure 10 wheel turns and divide the total to get a more accurate figure. If you do it right and measure accurately then you should be able to get it down to an error of approx 1% in the measurment, which will be the same error as you get in the distance. As said I'd be inclined to think the GPS is more out than the computer, especially over a shorter distance. Try a longer ride of several miles, on as straight a road as possible and see what difference they give.