Re:Pessimism about parallelism
by phkahler on Wednesday December 26, 2018 @08:56AM
(#18762136)
>>We can sum this up with a heuristic: Your odds of being able to apply parallel-computing techniques to a problem are inversely proportional to the degree of irreducible semantic nonlocality in your input data.
That's only true at the very high end, where you have so much data that it's not all physically close and transport time is relevant.
With the reality of stalled clock speeds and multi-core systems, people are finding the parallelism in many applications. Scaling really appears to be coming to an end - if not at 7nm then at 5 (or what they'll call 5). I don't think we're going to meaningfully exhaust our ability to exploit parallelism before then. They're not used in every application but they are used by enough things that we should have them.
That's only true at the very high end, where you have so much data that it's not all physically close and transport time is relevant.
With the reality of stalled clock speeds and multi-core systems, people are finding the parallelism in many applications. Scaling really appears to be coming to an end - if not at 7nm then at 5 (or what they'll call 5). I don't think we're going to meaningfully exhaust our ability to exploit parallelism before then. They're not used in every application but they are used by enough things that we should have them.