Page 1 of 1

Re: The computer race. [URL]

Posted: Mon Jun 26, 2017 3:26 pm
by bruce
That comparison may be based on CPU cycles only. What about those systems that use a massive number of GPUs?

"Operations" may mean standard (fixed-point) operations or it may include floating-point. in fact, do the count one AVX instruction as one or do they subdivide it into parallel sub-ops?

GPUs are typically rated by GFLOPS but they make it sound good by counting the speed of pure FMA operations. I suppose there's the same type of bias in the evaluation of supercomputers.

Re: The computer race. [URL]

Posted: Tue Jun 27, 2017 11:41 am
by Ricky
Here is an article on quantum computers that I found interesting. I don't think they will be folding anytime soon.
http://www.lanl.gov/discover/publicatio ... uantum.php

Re: The computer race. [URL]

Posted: Tue Jun 27, 2017 6:48 pm
by bruce
Riky: Thanks for the link. That's one of the clearest layman's explanation of quantum computers that I've seen.

FAH is solving an energy minimization problem and the golf ball / golf course example they give describes it extremely well. FAH starts a project by dropping a lot of golf balls and looks for areas where they collect. Then the formulate a new problem with a bunch of golf balls (temporarily?) in each of the popular collection sites and the examine how many of them then tunnel to a lower energy location -- and by what path -- and calculate how popular each of those paths are.

Classically, the final minimum is known, but FAH isn't looking for that. They're looking for how and why some of the golf balls ended up "stuck" in some other (almost minimum) destination.

I think FAH would do well to perform some of its studies on a quantum computer, but since they're not available on home computers, the real question becomes a more detailed study of what parts of FAH are being done well using distributed computing and what parts should be ported for operation on a (rended?) quantum computer.