mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   Oxford/Princeton AVX benchmarks (https://www.mersenneforum.org/showthread.php?t=16858)

davieddy 2012-05-30 09:50

Oxford/Princeton AVX benchmarks
 
How much speedup over a Core2 has George (et al) achieved
in LL testing?

David

fivemack 2012-05-30 10:59

[url]http://mersenne.org/report_benchmarks/?exover=1&exbad=1&specific_cpu=4373214[/url]

is an Ivy Bridge,

[url]http://mersenne.org/report_benchmarks/?exover=1&exbad=1&specific_cpu=2150153[/url]

is a core2.


Overclocking causes a bit of confusion, but it looks as if a 4096k FFT takes 90ms on the Core2 and 25ms or so on the Ivy Bridge. So a bit over a factor three.

davieddy 2012-05-30 11:21

THX Tom.
But does AVX alter the size of FFT needed for a given exponent?

D

Dubslow 2012-05-30 19:55

[QUOTE=davieddy;300693]THX Tom.
But does AVX alter the size of FFT needed for a given exponent?

D[/QUOTE]

Not significantly.

davieddy 2012-05-31 11:24

To avoid comparing apples with oranges,
how much speedup has AVX made to a given Ivy Bridge?

(Sorry for any Oxymorons)

D

axn 2012-05-31 11:54

[QUOTE=davieddy;300822]To avoid comparing apples with oranges,
how much speedup has AVX made to a given Ivy Bridge?
[/QUOTE]

About 30% (+/-). However, the speed gains are less, when run on multiple cores, due to memory bandwidth bottleneck.

davieddy 2012-05-31 19:36

[QUOTE=axn;300824]due to memory bandwidth bottleneck.[/QUOTE]
Has "memory contention" any relation to AVX?

BTW Don't you agree "bandwidth" is an overworked term?

axn 2012-06-01 03:11

[QUOTE=davieddy;300871]Has "memory contention" any relation to AVX?[/quote]
Don't know what you mean by "memory contention". Googling doesn't help.
[QUOTE=davieddy;300871]BTW Don't you agree "bandwidth" is an overworked term?[/QUOTE]
Dunno. However, in this context, it is not.

davieddy 2012-06-01 10:14

[QUOTE=axn;300910]Don't know what you mean by "memory contention". Googling doesn't help.

Dunno. However, in this context, it is not.[/QUOTE]

Different cores (or whatever) contending (competing) for memory access.

"Bandwidth" is used in just about ALL contexts these days!

Try Googling "bone of contention",
or "I used to be a contender".

David

davieddy 2012-06-01 11:15

Terminology
 
When Shannon (dubbed by Turing a poser) asked
von Neumann (he of "no hidden variables" notoriety)
what he should call his new term, the reply was:
"Call it Entropy. After all nobody understands what that means".

I wonder decreasingly as to why Boltzmann topped himself.

David

Dubslow 2012-06-01 19:47

[QUOTE=davieddy;300935]Different cores (or whatever) contending (competing) for memory access.

"Bandwidth" is used in just about ALL contexts these days!
[/QUOTE]

Yes; they're too darned fast. With more bandwidth, there would be enough room for all four/whatever cores to talk to mem as necessary, so there wouldn't be any contention.


All times are UTC. The time now is 22:50.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.