mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   CPU is too slow for normal trial factoring (https://www.mersenneforum.org/showthread.php?t=11193)

Aapje 2008-12-24 23:25

CPU is too slow for normal trial factoring
 
I have an Athlon 64 X2 2.1Ghz running XP 32 bit. I just switched to v5. When getting new assignments it says:
[code]
CPU is too slow for normal trial factoring assignments.[/code]When I was running on v4, I also let Prime95 select the work that makes most sense, but I got LL tests. Why am I getting something different now? I understand that TF is more efficient on Athlons, so in itself that makes sense, but why can't I do 'normal' TF? Are the exponents so big that normal TF only makes sense on 64 bit?

garo 2008-12-25 01:05

Do you have the number of hours the CPU will work per day set to a very low value?

cheesehead 2008-12-25 05:17

[quote=Aapje;154970]I have an Athlon 64 X2 2.1Ghz running XP 32 bit. I just switched to v5. When getting new assignments it says:
[code]
CPU is too slow for normal trial factoring assignments.[/code]When I was running on v4, I also let Prime95 select the work that makes most sense, but I got LL tests. Why am I getting something different now?[/quote]Firstly, the V5 server criteria for assignment types are more complicated than the V4 criteria were.

Secondly, a thread from three years ago ("Comparing Intel and AMD", [URL]http://mersenneforum.org/showthread.php?t=4517[/URL]) explained that:

1) Athlons are much better than Pentia when TFing to 62 bits or less.

2) SSE2 instructions allow much faster TFing than pre-SSE2 instructions when going above 64 bits.

So, Pentia with SSE2 performed much better than pre-SSE2 Athlons above 64 bits, at the time of that thread. However, I presume all recent Athlons have SSE2, so that shouldn't matter in your case.

Could be a bug: maybe the V5 criteria don't distinguish between SSE2 Athlons and pre-SSE2 Athlons for TF assignments?

Xyzzy 2008-12-25 16:49

Maybe power management is reporting a slower speed than 2.1GHz?

Aapje 2008-12-25 22:18

I've added CpuSpeed=2100 to local.txt to make sure that the correct speed is used. The program expects to run a respectable 8 hours a day. The CPU supports SSE2 (and SSE3). The primenet report correctly identifies the CPU, so it doesn't seem to misidentify it as an old, pre-SSE2 Athlon.

All in all, I think that I should get regular TF's, not just those that will be checked to low limits.

cheesehead 2008-12-26 02:57

[quote=Aapje;155058]I've added CpuSpeed=2100 to local.txt to make sure that the correct speed is used. The program expects to run a respectable 8 hours a day.[/quote]So, from PrimeNet's point-of-view, that's the same as a 700 MHz system running 24 hours a day, which it considers "too slow for normal trial factoring assignments".

[quote]All in all, I think that I should get regular TF's, not just those that will be checked to low limits.[/quote](I agree.)

Where does "program expects to run a respectable 8 hours a day" come from? Change that to a disrespectable 24 for the purpose of getting assignments, then change it back to the respectable 8 if necessary for whatever reason it had before. It won't matter later that your assignments take three times as long as expected, as long as you report them within six months, I think.

PrimeNet's not perfect; sometimes you have to tell it a disrespectable lie to get it to do the right thing.

Aapje 2008-12-26 09:07

That seems to be it, I now get LL double checks. Thanks.

lycorn 2008-12-26 23:24

That´s an interesting point.
I have a Core 2 duo laptop running TF (TF is chosen as the preferred type of work). Every time it requests a new work unit it receives the same message ("too slow for normal TF assignments"), but then it gets the "normal" assignents (currently exponents in the 54M range, and various TF levels from 65 to 68 bits). So I think there´s something not quite right about the message.
Before changing the number of working hours to 24, what were the exponents you were getting? In the 54M or in the 107-108M range? The TF to low limits is currently being done in the latter range.

Aapje 2008-12-27 00:37

54M. It factors to (at least) 67 bits. So it seems that the message is wrong.

Prime95 2008-12-27 03:52

Well, the message is right, but the code wasn't forcing LMH assignments. That is fixed.

According to [url]http://www.mersenne.org/thresholds/[/url] you need a 1GHz P4 equivalent to get non-LMH assignments. That is silly. I'll change it to 200MHz.

Aapje 2008-12-30 23:20

[quote=Prime95;155242]According to [URL]http://www.mersenne.org/thresholds/[/URL] you need a 1GHz P4 equivalent to get non-LMH assignments. That is silly. I'll change it to 200MHz.[/quote]

Thanks. However, I could not find a link to that thresholds page anywhere in PrimeNet. Perhaps that link should be placed on the Computer Properties and Account-Level Work Type Preference pages, so people know what happens when they select 'do what makes sense'.


All times are UTC. The time now is 22:06.

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