M1061... - Page 3

thomasn · 2004-10-18, 07:01

Quote:

Originally Posted by ET_

Did you send your results to George? He's got to update his ECM pages too.

Luigi

I have not yet done this, I will do so after I do a couple of more curves.

Thomas

Xyzzy · 2004-10-18, 18:04

We could send him a note every time we (as a group) complete 1000 curves... Or maybe email him once we've completed this bit depth...

Xyzzy · 2004-10-18, 18:54

126 curves done... (B1=44000000 & B2=184367799127)

BTW, if you post your curves, make sure that you are not posting your cumulative total... For example, every time I post, I clear my log file...

geoff · 2004-10-19, 04:09

Has anyone done a curve at the 55 digit (B1=110 million) level to see how long it will take and how much memory is used? I don't have a machine with enough memory to do stage two without increasing the -k parameter.

Xyzzy · 2004-10-19, 05:26

Quote:

Originally Posted by geoff

Has anyone done a curve at the 55 digit (B1=110 million) level to see how long it will take and how much memory is used? I don't have a machine with enough memory to do stage two without increasing the -k parameter.

Code:

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
 9809 mv        39  19  227m 222m 5252 R 93.9 45.0  46:10.69 ecm -c 0 -k 32 11e7

I used "-k 32" just as a guess... I have no idea how to set "k" properly... I can spare 300MB of memory so I guess I need to use a smaller "k"...

thomasn · 2004-10-19, 08:57

Quote:

Originally Posted by Xyzzy

We could send him a note every time we (as a group) complete 1000 curves... Or maybe email him once we've completed this bit depth...

Good Idea! I think mailing him every 1000 curves would seem like a correct frequency. We need to have the ecm web page updated every now and then so other do not duplicate our work.

BTW:
[Mon Oct 18 16:48:25 2004]
M1061 completed 100 ECM curves, B1=44000000, B2=4290000000

Who should keep the group curve count and e-mail George ?

Thomas

xilman · 2004-10-19, 14:24

Quote:

Originally Posted by Xyzzy

I have no idea how to set "k" properly... I can spare 300MB of memory so I guess I need to use a smaller "k"...

I have no idea how to set k properly either. What I do is set it as small as I can get away with and still have enough memory. Perhaps that is the "proper" approach.

Alex?

Paul

akruppa · 2004-10-19, 15:05

Memory use is, iirc, dF*(log_2(dF)+6), not sure about the 6, residues.

B2=B2min + d*dF*k, and d is approximately (well, not quite) proportional to dF.

Multiplying k by 4 therefore approx. halves dF, and so approx. halves memory use, for a given B2min-B2 interval. For a first approximation, memory use is a function of 1/sqrt(k).

Alex

Xyzzy · 2004-10-19, 23:41

I just upgraded from a 3200+ to a 3400+... AMD's naming scheme is a bit convoluted, so I'll describe the differences... The 3200+ I had was a 2GHz part with a 1024K L2... The 3400+ I have now is a 2.4GHz part with 512K L2...

Assuming the L2 cache size is irrelevant, this is a 20% jump in clock speed...

2000MHz/1024K

Code:

GMP-ECM 5.0.3 [powered by GMP 4.1.4] [ECM]
Input number is 24707306311927565716857342128774085333197833223161879682238935306082805123046306993647507776054336486228891340858985829027076261887914242781617846672453431386903982455635542158748401823985988322905245077938567513252198179128990807936780194781391547404884040101606295111368825026273254703636026307207764436438929167613951 (320 digits)
Using B1=44000000, B2=184367799127, polynomial Dickson(30), sigma=3552224530
Step 1 took 973900ms
Step 2 took 1134027ms

2400MHz/512K

Code:

GMP-ECM 5.0.3 [powered by GMP 4.1.4] [ECM]
Input number is 24707306311927565716857342128774085333197833223161879682238935306082805123046306993647507776054336486228891340858985829027076261887914242781617846672453431386903982455635542158748401823985988322905245077938567513252198179128990807936780194781391547404884040101606295111368825026273254703636026307207764436438929167613951 (320 digits)
Using B1=44000000, B2=184367799127, polynomial Dickson(30), sigma=2461182436
Step 1 took 812757ms
Step 2 took 965094ms

The overall difference is 18.6% faster...

Stage 1 = 19.8% faster...
Stage 2 = 17.5% faster...

I'm sure the fact I used different sigma values makes this less accurate... Anyways, in a nutshell, for ECM on the K8, a super large L2 cache isn't too important...

geoff · 2004-10-20, 01:04

Quote:

Originally Posted by Xyzzy

for ECM on the K8, a super large L2 cache isn't too important...

I think that as the exponent size is increased there will be a point where the number fits in 1024k but not in 512k cache, and so for a range of exponents the 3200+ might outperform the 3400+.

I noticed that the Celeron is comparable to the P4 for numbers up to M8192, but much slower for numbers above M32768, so somewhere in between is probably the point where the number no longer fits in a 128k cache.

Xyzzy · 2004-10-20, 01:19

Are we even dealing with FFTs here, like we are when we do GIMPS work? If so that would make a lot of sense... Hmm...

2004-10-19, 15:05	#30
akruppa "Nancy" Aug 2002 Alexandria 2,467 Posts	Memory use is, iirc, dF(log_2(dF)+6), not sure about the 6, residues. B2=B2min + ddFk, and d is approximately (well, not quite) proportional to dF. Multiplying k by 4 therefore approx. halves dF, and so approx. halves memory use, for a given B2min-B2 interval. For a first approximation, memory use is a function of 1/sqrt(k). Alex Last fiddled with by akruppa on 2004-10-19 at 15:08*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Predict the number of digits from within the factor for M1061	Raman	Cunningham Tables	12	2013-06-17 21:21
M1061 factored!!!	lycorn	NFS@Home	28	2012-08-30 04:40
Anyone have an ETA for M1061?	Stargate38	NFS@Home	99	2012-08-05 09:38
M1061 - t60	Andi47	Factoring	122	2011-11-25 09:18
P-1 on M1061 and HP49.99	ATH	Factoring	21	2009-10-13 13:16

2004-10-18, 18:04	#24
Xyzzy Aug 2002 8658₁₀ Posts	We could send him a note every time we (as a group) complete 1000 curves... Or maybe email him once we've completed this bit depth...

2004-10-18, 18:54	#25
Xyzzy Aug 2002 20722₈ Posts	126 curves done... (B1=44000000 & B2=184367799127) BTW, if you post your curves, make sure that you are not posting your cumulative total... For example, every time I post, I clear my log file...

2004-10-19, 04:09	#26
geoff Mar 2003 New Zealand 13·89 Posts	Has anyone done a curve at the 55 digit (B1=110 million) level to see how long it will take and how much memory is used? I don't have a machine with enough memory to do stage two without increasing the -k parameter.

2004-10-20, 01:19	#33
Xyzzy Aug 2002 2×3²×13×37 Posts	Are we even dealing with FFTs here, like we are when we do GIMPS work? If so that would make a lot of sense... Hmm...