Intel Threading Building Blocks

What did they really say?

Here is what Amdahl and Gustafson actually said in their famous papers, which have generated much dialog ever since:

…the effort expended on achieving high parallel processing rates is wasted unless it is accompanied by achievements in sequential processing rates of very nearly the same magnitude.—Amdahl, 1967

…speedup should be measured by scaling the problem to the number of processors, not by fixing the problem size.—Gustafson, 1988

Combining these ideas, you can conclude that the value of parallelism is easier to prove if you are looking forward than if you assume the world is not changing. If we double the number of cores every couple of years, we should be working to double the amount of work we want our computer to do every couple of years as well.