Amdahl's Law Is Misleading

embeddedrelated.com

45 points·chewxy··13 comments

This scatter-brained exposé really says nothing about Amdahl's Law except that there are still benefits to optimizing pieces that are not necessarily the main "bottleneck" of a task, which I think is obvious in relevant contexts.

Amdahl's law is not misleading. Amdahl's law shows the speed up under a fixed problem size. Gustafson simple showed that since we parallelized an aspect of a problem, we can solve larger problems with the time saved. Amdahl's law is the max for fixed problem sizes.

Reevaluating Amdahl's Law.

http://www.johngustafson.net/pubs/pub13/amdahl.pdf

wow even this is a bit hard to parse (and I work with John!)

Basically, Gustafson's law says, "yeah but really what you're going to do with massively parallel computing is you're going to run more copies of the same program, which has a linear scaling factor".

The misleading part is how it is used. As things change, theoretical speed up of the solution becomes irrelevant, and problems also change.

Lol, seeing the comments so far I might be the minority, but I loved reading this. Made my mind feel like it came up from under water to see the sun. Happy

How does amdahls law apply to articles that take two hours to read?

It says that you don't parallelize well.

That's not really fair. I skimmed it in less than 5 minutes, and I thought it was worthwhile.

The title is really awesome!

Wow, this was really not worth the time it took to read it.

Reply on news.ycombinator.com