mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   P-1 factoring anyone? (https://www.mersenneforum.org/showthread.php?t=11101)

James Heinrich 2011-04-13 17:14

For my part, I've got 10 cores running P-1:
4x i7-920 @ 3.2GHz w/12GB (10GB allocated)
4x Q6600 @ 2.7GHz w/8GB (6GB allocated)
2x 3600X2 @ 2.0GHz w/2GB (1GB allocated)

The older machines are working their way through ancient ranges of never-really-P1'd exponenents (and finding a bunch of factors, I might add). My i7 is working through more current exponents. The i7 has plenty of bandwith available (triple-channel DDR3-1600) and is unbothered by which exponents are in stage1/2. The Q6600 is dual-channel DDR3-1333, and is only slightly bothered. (The 3600+ is too slow for me to really pay any attention to.) On both my quads, I limit Stage2 to max 3 workers at a time, but more to give each one a decent chunk of RAM rather than for performance bottleneck reasons.

petrw1 2011-04-13 17:46

[QUOTE=James Heinrich;258426] On both my quads, I limit Stage2 to max 3 workers at a time, but more to give each one a decent chunk of RAM rather than for performance bottleneck reasons.[/QUOTE]

And do you use the "Memory=" parms to limit Memory use per core or do you let it use whatever is available? (i.e. if only 1 core is in stage 2 it gets all the RAM; if there are 2 cores then half each and if 3 cores then 1/3 each)

James Heinrich 2011-04-13 17:55

I let them fight it out for memory. The Q6600 is currently working in the M7 range, so a full batch of 480 relative primes only takes about 1500MB, so there's no real fighting for RAM on that machine. On the i7, it varies depending on workload. Sometimes I'll see workers with 6GB/2GB/1GB, other times it's balanced at 3/3/3. Of course, the win comes when only 2 or 1 worker happens to be in stage2 at that time, in which case they get a bunch more RAM than if I hardcoded a per-worker limit. Right now I have a big backlog of stage2 to complete, so it's rare to see less than the max workers in stage2.

science_man_88 2011-04-13 18:21

[QUOTE=petrw1;258412]Depends completely on what CPU you are running on and how much RAM you allocate.

I have a i5-750 that will complete a P-1 in the 50M range in 31 hours and a PIV 2.4 Ghz that takes over 100 hours.[/QUOTE]

yeah I've said it before but I'm on a HP comnpaq presario SR2050NX ( apparently designed for XP)with a 2.80 Ghz processor 3.19 GB of detected ( usable ?) ram on a 32 bit installation of windows 7 and a ATI graphics with almost all USB ports filled with plug in stuff and a headset if I want to do speech recognition. I have a new printer a energy star model of almost the same one I used to have I think ( 3 in one I believe). At last check it got a 2.2 window experience rating ( but that's because it sucks at aero).

petrw1 2011-04-13 19:10

[QUOTE=science_man_88;258434]yeah I've said it before but I'm on a HP comnpaq presario SR2050NX ( apparently designed for XP)with a 2.80 Ghz processor 3.19 GB of detected ( usable ?) ram on a 32 bit installation of windows 7 [/QUOTE]

I tkink this is really all that is relevant to P-1 times.

With this setup I don't see why it should take more than about 4 days for P-1 in the current range assuming you allocate at least 400MB to it and run 24 hours a day.

petrw1 2011-04-13 19:14

[QUOTE=James Heinrich;258432]I let them fight it out for memory.[/QUOTE]
One more question... I know that it chooses B1/B2 based partly on how much RAM it has to work with. If I were to allocate 2400M it will choose larger B1/B2 than if I allocate 600M. But if I allocated 2400M total and 600M for each of the 4 cores will it not choose smaller B1/B2 and potentially complete each one faster?

James Heinrich 2011-04-13 19:38

[QUOTE=petrw1;258440]With this setup I don't see why it should take more than about 4 days for P-1 in the current range assuming you allocate at least 400MB to it and run 24 hours a day.[/QUOTE]Expounding a little: The CPU is a [url=http://ark.intel.com/Product.aspx?id=27512]Pentium D 820[/url] dual-core (2.8GHz, 2x 1MB cache). A single core should push through about [url=http://mersenne-aries.sili.net/throughput.php?cpu1=Intel%28R%29+Pentium%28R%29+D+CPU+2.80GHz|1024|0&mhz1=2800]1.58GHz-days per day[/url] of P-1 work in the 53M range. If a single P-1 assignment is about 4GHz-days of work then a single core should really only take about 2.5-3 days to complete. Don't expect that running two P-1's on that CPU to give you double the throughput, but you should be able to complete a P-1 at least every other day, I'd say. Assuming you have a decent amount of RAM allocated, that is (you didn't say what amount of RAM you let Prime95 use...?)

science_man_88 2011-04-13 20:00

[QUOTE=James Heinrich;258447]Expounding a little: The CPU is a [url=http://ark.intel.com/Product.aspx?id=27512]Pentium D 820[/url] dual-core (2.8GHz, 2x 1MB cache). A single core should push through about [url=http://mersenne-aries.sili.net/throughput.php?cpu1=Intel%28R%29+Pentium%28R%29+D+CPU+2.80GHz|1024|0&mhz1=2800]1.58GHz-days per day[/url] of P-1 work in the 53M range. If a single P-1 assignment is about 4GHz-days of work then a single core should really only take about 2.5-3 days to complete. Don't expect that running two P-1's on that CPU to give you double the throughput, but you should be able to complete a P-1 at least every other day, I'd say. Assuming you have a decent amount of RAM allocated, that is (you didn't say what amount of RAM you let Prime95 use...?)[/QUOTE]

yeah well I had it running full tilt ( though maybe not without sleep mode ( which may be the problem)) with 2936 ( the maximum it allowed) and at one day I think it was under 4%.

petrw1 2011-04-13 20:27

[QUOTE=science_man_88;258451]yeah well I had it running full tilt ( though maybe not without sleep mode ( which may be the problem)) with 2936 ( the maximum it allowed) and at one day I think it was under 4%.[/QUOTE]

Do you happen to recall if it was 4% of stage 1 or stage 2?

Typically Stage 2 takes about 50% longer than Stage 1 so if by any change it was 4% of Stage 2 then it was really almost half done.

science_man_88 2011-04-13 20:28

[QUOTE=science_man_88;258451]yeah well I had it running full tilt ( though maybe not without sleep mode ( which may be the problem)) with 2936 ( the maximum it allowed) and at one day I think it was under 4%.[/QUOTE]

never mind apparently I had my preferred settings on my PC set to balanced performance. It's now on high performance. that may be a factor.

James Heinrich 2011-04-13 21:50

It sounds very much like it was at 4% of stage 2, so in reality almost exactly 50% done. If you would've left it for another day it would've completed.


All times are UTC. The time now is 23:04.

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