Previous | Next --- Slide 40 of 51
Back to Lecture Thumbnails
BestBunny

What are some example scenarios in which a relatively higher performance on conjugate gradient is better than a lot higher performance for LINPACK? If LINPACK performance is a lot better than conjugate gradient, can't conjugate gradient problems be converted to LINPACK?

vasua

I can't seem to find a source for this, but I remember hearing that one of the top supercomputers (I think it's this one) was designed and optimized primarily to beat benchmarks, rather than to be particularly useful in real world applications. While I'm sure that there's tons of use for 3 million cores, it's a bit disheartening to realize that shady things like this happen at all levels (assuming the rumor is true).