I'd be surprised to see a 45nm chip from AMD in quantity before the end of this year.
my understanding is that it's actually an iterative algorithm that takes 19 iterations to converge to the number of digits required. it's not just streaming out of cache.Could also be caching hierarchy is better; the current line of chips has terrible latency issues on the L3. A 1M digits of Pi has a far better chance of being cache coherent on 6mb of shared cache, and if the timings are better, that could certainly explain the performance in that particular benchmark.
More benches are needed to determine how much better the computational efficiency is, not just for 19 successive calls of the same Pi calculation... Just IMO of course.
my understanding is that it's actually an iterative algorithm that takes 19 iterations to converge to the number of digits required. it's not just streaming out of cache.
http://en.wikipedia.org/wiki/Gauss–Legendre_algorithm
SuperPi 1MB
===========
Conroe @ 4ghz: 12,85 s
Allendale @ 4ghz: 14,47 s
Conroe @ 3ghz: 17,13 s
Allendale @ 3 ghz: 19.29 s
Difference: Conroe 12.6% faster
SuperPi 32MB
===========
Conroe @ 4ghz: 13m 02.40 s
Allendale @ 4ghz: 17m 21.00 s
Conroe @ 3ghz: 17m 23.04 s
Allendale @ 3 ghz: 23m 07.80 s
Difference: Conroe 33.1% faster
well sure, some stuff is going to be in the cache, but it's not just streaming out of cache over and over and over. it's not like you're running the same exact computation with the same exact inputs 20 times.
Me said:I'm not saying that's the entire reason it's going faster, I'm not even going to say it's certain to be relevant, but I am saying that the SuperPi result is dubious at best in terms of what it can tell us about the performance of this new processor.
Yorkfield is also very good in that department though:Still, for someone wanting Q6600-like performance at very low power, this could be your ticket. Nothing wrong with that stance at all; more and more people are looking for ways to save energy.
Yorkfield is also very good in that department though:
http://www.behardware.com/articles/720-2/test-six-x3-x4-phenom-b3s-in-detail.html
Sorry but those scores are very dissapointing. My year and a half old E4300 @ 3,2GHz (400MHz FSB) takes 18 seconds for 1M test... though i indeed only have just 2 cores altogether But still...
I really wish AMD to make a huge comeback with something of Athlon4, AthlonXP or Athlon64 style...
Those were kickin' Intels butt badly in those days.