mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   3*2^n-1 Search (https://www.mersenneforum.org/forumdisplay.php?f=14)
-   -   HT processors (https://www.mersenneforum.org/showthread.php?t=6361)

paulunderwood 2006-09-20 18:30

HT processors
 
Thanks to our latest member, arminius, it has been shown that running two instance of LLR (321) improves throughput. All you need to do is create a new directory/folder (called llr2 or whatever) and run the second instance from there. You will have to have a HT enabled processor and motherboard (and SMP kernel if linux.)

I am unsure about a 533FSB motherboard which is HT but with a 533FSB northwood. It is upto you to do your own tests...

This might help:
[url]http://en.wikipedia.org/wiki/Pentium_4[/url]
[url]http://en.wikipedia.org/wiki/Pentium_D[/url]

Good luck! :grin:

ET_ 2006-09-21 14:03

[QUOTE=paulunderwood;87585]Thanks to our latest member, arminius, it has been shown that running two instance of LLR (321) improves throughput. All you need to do is create a new directory/folder (called llr2 or whatever) and run the second instance from there. You will have to have a HT enabled processor and motherboard (and SMP kernel if linux.)

I am unsure about a 533FSB motherboard which is HT but with a 533FSB northwood. It is upto you to do your own tests...

This might help:
[url]http://en.wikipedia.org/wiki/Pentium_4[/url]
[url]http://en.wikipedia.org/wiki/Pentium_D[/url]

Good luck! :grin:[/QUOTE]

wHAT ABOUT amd x2? :question:
Luigi

paulunderwood 2006-09-21 15:03

Presumably, with a [I]dual-core HT[/I] chip, you could run 4 instances for maximum throughput. I don't have such a chip. For testing, if you create 4 (321) LLR directories with exactly the same test numbers in each, then check 2 instances running and see if the iteration time is less than double when 4 instances are running (at all times letting the iteration times settle down at say 1000 iterations per screen output.) :unsure:

paulunderwood 2006-09-21 15:28

Here is are results for 3.4Ghz Northwood:

single LLR:
3*2^2827583-1 is not prime. LLR Res64: AFC6CD0A3099B096 Time : 10972.677 sec.

two LLRs:
3*2^2827658-1 is not prime. LLR Res64: FD4EBE816264EF1C Time : 20796.945 sec.

Allowing for a slight variation because these numbers are not exactly the same, that's:
? 2*10972.677/20796.945
1.055220081603331643181246091

a 5.5% gain.

(I wish I had realised this a few years back :blush: )

arminius 2006-09-21 15:58

[QUOTE=paulunderwood;87669]
(I wish I had realised this a few years back :blush: )[/QUOTE]

New members - new ideas :rolleyes: Seriously, I'm glad to see thats working.
a.

ET_ 2007-02-15 13:07

I did some tests.

I have a Prescott Pentium4 @3,2 GHz and an AMD64 X2 @ 2.0 GHz, both with 1 GB of RAM.

I am running a copy of Prime95 and a copy of LLR 3.7.0 on both machines.

Both fftlen are 163840.

The timing on the AMD64 X2 is exactly ONE HALF the timing of the Pentium4: LLR runs TWICE as fast on AMD64 X2 on the same configuration with another instance of Prime95 running.

Luigi :et_:

Thomas11 2007-02-15 13:34

Luigi, I'm not sure whether I fully understand your results.

If I remember correctly, the Prescott is a single-core HT cpu while the Athlon64 X2 is a true dual-core cpu. Therefore you could run two instances of LLR or Prime95 with almost no impact to each other on the Athlon64 X2. But if you do the same on a single-core (HT or not HT doesn't matter really much), each instance would run only at half the speed...

ET_ 2007-02-15 15:47

[QUOTE=Thomas11;98594]Luigi, I'm not sure whether I fully understand your results.

If I remember correctly, the Prescott is a single-core HT cpu while the Athlon64 X2 is a true dual-core cpu. Therefore you could run two instances of LLR or Prime95 with almost no impact to each other on the Athlon64 X2. But if you do the same on a single-core (HT or not HT doesn't matter really much), each instance would run only at half the speed...[/QUOTE]

I lost my mind somewhere else, sorry...

Hyperthreaded CPUs and dual core ones show equal notation on task manager, and I had switched off my brain before writing.

Oh, well shat hippens... :blush: :blush:

Luigi


All times are UTC. The time now is 08:54.

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