Hacker News Reader

Of course the real difference is a lot higher as it's not just about the clock rate - an awful lot of improvements were made in areas such as the pipeline, caching, floating-point support etc. so I wouldn't be very surprised if the real performance ratio was higher than 1000.

View on HN

Floating point ops used to take a while. A 16Mhz Amiga with a floating point co-processor might have had one of these:

http://en.wikipedia.org/wiki/Motorola_68881

Yeah, 16 Mhz ==> 160 KFlops, if you believe wikipedia. Supposedly the i7 has a peak of 79 GFlops. Sure, you can't get all of that performance out (either then, or now), but 79,000,000 / 160 is a bit larger than 143.

You used to measure the number of cycles an integer multiply costed, in the dozens or hundreds. Now it takes 1/4 of a cycle. A factor of 10,000 might be a fairer number

Reply on news.ycombinator.com