Minimum/desired CPU specs for ECM factoring

Kaboom · 2009-04-09, 11:50

I have a couple low-spec PCs in my "cluster" which are currently doing PM1-L assignments. I was wondering whether it would be better to switch them to ECM or ECM-F.

Specs are thus:
Athlon XP 1800+ (1533 MHz clock)
512 MB RAM
O/S is Windows XP Pro on one and Gentoo Linux on the other.

Which of these three assignment preferences are better for the hardware (I'd rather avoid TF)?

Thanks in advance for comments.

ET_ · 2009-04-09, 12:47

Quote:

Originally Posted by Kaboom

I have a couple low-spec PCs in my "cluster" which are currently doing PM1-L assignments. I was wondering whether it would be better to switch them to ECM or ECM-F.

Specs are thus:
Athlon XP 1800+ (1533 MHz clock)
512 MB RAM
O/S is Windows XP Pro on one and Gentoo Linux on the other.

Which of these three assignment preferences are better for the hardware (I'd rather avoid TF)?

Thanks in advance for comments.

Geore redefined Fermat ECM, giving out numbers based on the requested/available memory, so I'd choose ECM on Fermat numers.

Luigi

petrw1 · 2009-04-09, 14:52

Quote:

Originally Posted by Kaboom

I have a couple low-spec PCs in my "cluster" which are currently doing PM1-L assignments. I was wondering whether it would be better to switch them to ECM or ECM-F.

Per the "P-1 Factoring Anyone" thread I think they need you more than ECM does.

James Heinrich · 2009-04-10, 12:14

But with only 512MB RAM in the entire system it's on the low end of specs for current P-1 assignment. However, as long as you have as much RAM as practical dedicated to Prime (e.g. at least 200MB) then you can make a useful contribution to ECM-M, ECM-F or P-1.
P-1 is "most useful" to the main GIMPS project.
ECM-F is most tuned to available RAM on the client computer.
ECM-M also has relatively-low RAM requirements.

James Heinrich · 2009-04-10, 14:15

Correction: both ECM-F and ECM-M are equally "tuned" to the amount of RAM available.

Prime95 · 2009-04-10, 14:38

Quote:

Originally Posted by James Heinrich

But with only 512MB RAM in the entire system it's on the low end of specs for current P-1 assignment.

That is true -- but he will have more RAM available than LL testers that have left the memory setting at the default 8MB. If we don't get more P-1 factorers, then some of the P-1 runs will be done by these 8MB LL testers.

James Heinrich · 2009-04-10, 15:22

I heartily agree that anyone willing to do P-1 with a half-decent amount of RAM should be encouraged to do so (including, per the other thread, clients set to Whatever Makes Sense with a generous amount of RAM).

Kaboom · 2009-04-14, 06:38

Thanks to everyone for the replies. All considered I think I'll keep doing P-1 tests, then.

petrw1 · 2009-04-16, 15:26

I just completed my first ECM-F on 1 core of a Q9500.
About a month ago it completed a ECM-M test.

However, the points vs the time taken is lower than other assignment types. Is this by design?

I have 1200M Allocated. The other 3 cores were doing 28M LL tests.

Quote:

8388608 NF-ECM 2009-04-16 09:35 14.8 3 curves, B1=250000, B2=25000000 4.1562

It took 69 hours to complete which works out to 1.44 points per day.

A Mersenne ECM assignment on this same Quad (albeit a small one) computes to the same 1.44 points per day.

Quote:

2741183 NF-ECM 2009-03-18 10:19 7.3 3 curves, B1=50000, B2=5000000 0.2426

This one took 4 hours 2 minutes.

This same core completes a 28M LL in 12 days for an average of about 2.3 points per day ... 50% more.

James Heinrich · 2009-04-16, 23:59

My guess would be your system is starved for RAM bandwidth. I'll guess that you have 800MHz RAM in there? I bet if (as a test) you ran a single ECM (one core, other 3 cores idle) you'd get closer to the expected throughput (my as-yet-unreleased throughput calculator estimates about 2.81GHz-days/day for a single core of the Q9550 @ 2.83GHz).

petrw1 · 2009-04-17, 14:58

Quote:

Originally Posted by James Heinrich

My guess would be your system is starved for RAM bandwidth. I'll guess that you have 800MHz RAM in there? I bet if (as a test) you ran a single ECM (one core, other 3 cores idle) you'd get closer to the expected throughput (my as-yet-unreleased throughput calculator estimates about 2.81GHz-days/day for a single core of the Q9550 @ 2.83GHz).

I could try the single core test, I know that even my LL iterations times drop from 0.036 to 0.028 for 28M exponents with only 1 core running.

However as for the RAM I specifically ordered DDR2 1066 RAM; 4 Gb of it.

2009-04-09, 11:50	#1
Kaboom Apr 2003 Milan, Italy 2×3×5 Posts	Minimum/desired CPU specs for ECM factoring I have a couple low-spec PCs in my "cluster" which are currently doing PM1-L assignments. I was wondering whether it would be better to switch them to ECM or ECM-F. Specs are thus: Athlon XP 1800+ (1533 MHz clock) 512 MB RAM O/S is Windows XP Pro on one and Gentoo Linux on the other. Which of these three assignment preferences are better for the hardware (I'd rather avoid TF)? Thanks in advance for comments.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Not getting desired assignment type	Chuck	GPU to 72	2	2013-01-13 16:32
New hardware specs needed	ET_	Hardware	4	2009-12-21 15:19
Computer Specs	Unregistered	Information & Answers	16	2009-01-03 07:36
your time/iter and specs?	ixfd64	Hardware	0	2005-12-13 11:40
Desired Features/Addins	Complex33	Software	5	2002-09-05 06:13

2009-04-10, 12:14	#4
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 110011000110₂ Posts	But with only 512MB RAM in the entire system it's on the low end of specs for current P-1 assignment. However, as long as you have as much RAM as practical dedicated to Prime (e.g. at least 200MB) then you can make a useful contribution to ECM-M, ECM-F or P-1. P-1 is "most useful" to the main GIMPS project. ECM-F is most tuned to available RAM on the client computer. ECM-M also has relatively-low RAM requirements.

2009-04-10, 14:15	#5
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 2×3×5×109 Posts	Correction: both ECM-F and ECM-M are equally "tuned" to the amount of RAM available.

2009-04-10, 15:22	#7
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 2·3·5·109 Posts	I heartily agree that anyone willing to do P-1 with a half-decent amount of RAM should be encouraged to do so (including, per the other thread, clients set to Whatever Makes Sense with a generous amount of RAM).

2009-04-14, 06:38	#8
Kaboom Apr 2003 Milan, Italy 2·3·5 Posts	Thanks to everyone for the replies. All considered I think I'll keep doing P-1 tests, then.

2009-04-16, 23:59	#10
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 2×3×5×109 Posts	My guess would be your system is starved for RAM bandwidth. I'll guess that you have 800MHz RAM in there? I bet if (as a test) you ran a single ECM (one core, other 3 cores idle) you'd get closer to the expected throughput (my as-yet-unreleased throughput calculator estimates about 2.81GHz-days/day for a single core of the Q9550 @ 2.83GHz).