mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Riesel Prime Search (https://www.mersenneforum.org/forumdisplay.php?f=59)
-   -   5*2^n-1 Reservation Thread (https://www.mersenneforum.org/showthread.php?t=9472)

pinhodecarlos 2013-11-29 16:59

I am very disappointed by the Intel® Xeon® Processor E5-2630 v2 performance for k=5 at n=3880k range. It is almost 800 seconds slower than my laptop (Toshiba Satellite L850-1HZ with Intel® Core™ i7-3630QM Quad Core (Ivy Bridge) ).

amphoria 2013-11-29 18:04

If you are using all 6 cores than you are almost certainly saturating the memory bus. I did some testing with an E5-2680 which has similar memory bandwidth. With up to 4 cores there was no slowdown. With 6 cores the slowdown was so much that the overall throughput was less than with only 4 cores. It was only with 8 cores that the extra cores made up for the bandwidth constraints and the total throughput was greater than with 4 cores, ie. the slowdown is not linear once you exceed 4 cores.

Note that my testing was with 80K FFTs. I don't whether the larger FFT size you will be using is likely to also exceed the 2.5 MB L2 cache per core.

pinhodecarlos 2013-12-02 10:13

1 Attachment(s)
3726-3746 complete, no primes.

pinhodecarlos 2013-12-02 10:20

[QUOTE=amphoria;360638]If you are using all 6 cores than you are almost certainly saturating the memory bus. I did some testing with an E5-2680 which has similar memory bandwidth. With up to 4 cores there was no slowdown. With 6 cores the slowdown was so much that the overall throughput was less than with only 4 cores. It was only with 8 cores that the extra cores made up for the bandwidth constraints and the total throughput was greater than with 4 cores, ie. the slowdown is not linear once you exceed 4 cores.

Note that my testing was with 80K FFTs. I don't whether the larger FFT size you will be using is likely to also exceed the 2.5 MB L2 cache per core.[/QUOTE]

One core is used while the others are doing some DNS simulations (CFD simulations).

pinhodecarlos 2013-12-03 10:03

1 Attachment(s)
3930-3940 complete, no primes.

pinhodecarlos 2013-12-03 11:03

1 Attachment(s)
Let's update first post with:

[B]The latest test file[/B]
The file contains all candidates in the 4-5M range.
Please help yourself by extracting the range you want to test.
Extensively sieved by Psieve (latest update of November 30, 2013 ~337P)
[B]Link to file here[/B] (35016 candidates)

From my side I have now 8 cores on my current k=5 ranges and then I will add more 4 as my 1st and 2nd megabit drive ranges finishes so I can dedicate all cores to k=5 for next year.

Carlos

Kosmaj 2013-12-03 22:53

The first post is updated!

And I'm taking 4000-4010.

pinhodecarlos 2013-12-04 11:06

[QUOTE=Kosmaj;361086]
And I'm taking 4000-4010.[/QUOTE]

Damm you...lol Good luck.

pinhodecarlos 2013-12-05 11:38

[B]The latest test file[/B]
The file contains all candidates in the 3-4M range, including already tested ones.
Please help yourself by extracting the range you want to test.

should be

[B]The latest test file[/B]
The file contains all candidates in the 4-5M range, including already tested ones.
Please help yourself by extracting the range you want to test.

pinhodecarlos 2013-12-16 14:12

Taking 4010-4090.
(Just queuing up more work due to the fact that I will be away from the computers)

pinhodecarlos 2013-12-17 10:19

1 Attachment(s)
3756-3766 complete, no primes.


All times are UTC. The time now is 22:48.

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