"bearophile" <[email protected]> wrote in message news:[email protected]... > Justin Johansson: >> "I think we have maybe five to 10 years left [with Moore's Law]," he >> said. "Optimizations will get very, very sexy again, when people realize >> how we pay for abstractions." > > We'll see, but I don't believe that. We'll see. CPUs with 30 cores require > a different kind of optimizations. >
First of all, when we get up around 30 cores (more or less), we'll be getting right back into the same problem we had a few years back with rediculous power consumption and heat generation. Only this time, there will be fewer things that can be done about it. But ignoring that... Fast-forward to that day when there are 30-core CPUs, and grab three programs: A. Something optimized for those 30-cores. B. Something optimized with traditional techniques. C. Something like Bazaar. 1. Certainly, A is going to run fastest...except perhaps if the user is simultaneously running five other programs that are also optimized for 30 cores each...or if the user isn't running on a fancy top-of-the-line 30-core machine and is only getting by with a cost-effective 8 cores with is more than sufficient for their needs...in which cases B can quite realistically compete with A. 2. B is still going to kick the shit out of C. 3. Grab a fourth program, that has both trditional optimizations and 30-core optimizations, provided that they are both balanced in an intelligent way, and that'll probably beat them all, demonstrating that 30-core optimizations are be something to suppliment, not replace, traditional optimizations.
