Anybody want to explain how the median on the figure below is 1350? I grabbed this figure from an article on CUDA vs OpenCL and who will win for parallel computing. As far as I can tell, it has rank on the x-axis and speed-up on the y-axis. Since there are 100 observations, the median will be the average of the 50th and 51st observations, which both look to be around 300. I tried to calculate a mean giving the benefit of the doubt to all speed-ups and I only got 605. How could they possibly get 1350? Separately, in the top 100 NVIDIA showcases, the last 40 or so are under 200-fold speed-up? In my applications, I have seen 200-fold speed-ups. Am I on the list? I'm guessing not, so how do I get on the list?

blog comments powered by Disqus


06 October 2011