Core2 X6800 Test Times

PrimeCrazzy · 2006-08-20, 04:58

Based on the benchmarks it will take a Core2 X6800 2.9 MH about 17 days to LL test a 35,000,000 digit number. Can this computer test 2 numbers at once, thus testing 2 numbers in 17 days?

sdbardwick · 2006-08-20, 05:13

It will probably be more like 2 numbers in 18.5 days. Each number can run on its own core, but one cannot expect 100% linear scaling.

BlueCatZ1 · 2006-08-21, 20:50

On a X6800 Conroe in XP MCE here are some numbers:

With one instance of P95 at 2048 fft I was getting .041 sec per iteration.

With one instance LL testing and one in P1 factoring I was getting .051 sec/iter.

With two instances of P95 at 2048 fft I was getting .046 seconds/iteration.

So 16.9 Days for One 10M digit LL test or 18.97 Days for two? Do these numbers sound right? Hope this helps.

-Zak

PrimeCrazzy · 2006-08-24, 18:07

Those numbers sound great. I currently am running a P4 at 2.9Mh. I need to convince my wife it is time for an upgrade.

Mystwalker · 2006-08-24, 19:03

I would be interested in gmp-ecm benchmarks of the Core2.

Does anybody care to run some curves with it?

BlueCatZ1 · 2006-08-24, 20:49

Quote:

Originally Posted by Mystwalker

I would be interested in gmp-ecm benchmarks of the Core2.

Does anybody care to run some curves with it?

I would be happy to if you let me know what/how to do that. What numbers, amount of time etc.?

-Zak

Mystwalker · 2006-08-26, 16:42

Thanks for your offer, Zak!

First, download the binaries from here.
Unfortunately, there are no binaries for the Core2 yet. You could try those for the P4 (it should be quite irrelevant which specific, I guess) and/or the Athlon binaries.

Then create a text file in the same directory, with the following contents:
512697074397868507464830992105855089664575695425522835992747470881488640052598076608364916264171931518711016452728463454015615930496173963179151714060127379851269343356435328168074844330850160596351576082971234903468258958205452547120425573322600440367004964883936437287906385278156561312381803353061

Afterwards, execute the program as follows:
ecm 1000000 < [textFile]

Where [textFile] is the name of the text file. It should run for 2-4 minutes.
Please do this twice and report back the output.

Compiling the sources would be best, but involves quite some manual work. Since this instruction, GMP and gmp-ecm have improved, you'd need GMP 4.2 and gmp-ecm 6.1.1
If you want to try it, I'm confident you'll get enough feedback from the forum to solve any problems in your way.

BlueCatZ1 · 2006-08-28, 03:21

Here are results from ECM

(Version I was able to compile)(Was saying stuff for pentium3 durring install?)
GMP-ECM 6.1.1 [powered by GMP 4.2.1] [ECM]
Input number is
(300 digits)
Using B1=1000000, B2=1045563762, polynomial Dickson(6), sigma=2996848946
Step 1 took 47390ms
Step 2 took 18485ms

(Version from forums for Prescott p4)

GMP-ECM 6.1.1 [powered by GMP 4.2.1] [ECM]
Input number is
(300 digits)
Using B1=1000000, B2=1045563762, polynomial Dickson(6), sigma=1801303844
Step 1 took 41703ms
Step 2 took 17109ms

Tests done on Core 2 Duo X6800 with With 4 GB Ram.

Hope This helps.

-Zak

Mystwalker · 2006-08-28, 19:20

Thanks for the testing, Zak!

Even with binaries that aren't optimized for the CPU, it is 40% (Step1) to over 60% (Step2) faster than a P4 Prescott - at least for this first test.

I wonder how the Core2 performs in relation to an Athlon64. I guess that currently, the A64 will pull away (due to assembly support), but the Core2 could be a real competitioner once the software is optimized.

Dresdenboy · 2006-08-29, 08:34

Quote:

Originally Posted by Mystwalker

I wonder how the Core2 performs in relation to an Athlon64. I guess that currently, the A64 will pull away (due to assembly support), but the Core2 could be a real competitioner once the software is optimized.

Thanks to a architecture closer to A64 than to P4, the Core2 might also be in the lead in this test. It depends on the instruction mix. It's likely, that the K8 optimized assembler code will run on the Core2 - and in this case it will be doing fine... better than on a P4.

If the instruction mix is more SSE2, then Core2 will benefit. If it's more based on integer code, then we need to see results for a conclusion.

2006-08-20, 04:58	#1
PrimeCrazzy Dec 2005 E₁₆ Posts	Core2 X6800 Test Times Based on the benchmarks it will take a Core2 X6800 2.9 MH about 17 days to LL test a 35,000,000 digit number. Can this computer test 2 numbers at once, thus testing 2 numbers in 17 days?

2006-08-26, 16:42	#7
Mystwalker Jul 2004 Potsdam, Germany 3×277 Posts	Thanks for your offer, Zak! First, download the binaries from here. Unfortunately, there are no binaries for the Core2 yet. You could try those for the P4 (it should be quite irrelevant which specific, I guess) and/or the Athlon binaries. Then create a text file in the same directory, with the following contents: 512697074397868507464830992105855089664575695425522835992747470881488640052598076608364916264171931518711016452728463454015615930496173963179151714060127379851269343356435328168074844330850160596351576082971234903468258958205452547120425573322600440367004964883936437287906385278156561312381803353061 Afterwards, execute the program as follows: ecm 1000000 < [textFile] Where [textFile] is the name of the text file. It should run for 2-4 minutes. Please do this twice and report back the output. Compiling the sources would be best, but involves quite some manual work. Since this instruction, GMP and gmp-ecm have improved, you'd need GMP 4.2 and gmp-ecm 6.1.1 If you want to try it, I'm confident you'll get enough feedback from the forum to solve any problems in your way. Last fiddled with by Mystwalker on 2006-08-26 at 16:43

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Odd scaling of test times between two machines	mdettweiler	Hardware	3	2014-07-28 16:35
Running 50% util on i5 and Core2 (same MHz)	SO7783	Hardware	13	2010-03-01 22:59
Optimizing Core2 quad in Windows XP	John Rheinstein	Hardware	18	2009-09-23 16:14
Intel core2 Duo sieving?	cipher	Twin Prime Search	15	2007-06-05 21:20
Another Core2 Duo question	Ender	Hardware	3	2007-02-08 00:12

2006-08-20, 05:13	#2
sdbardwick Aug 2002 North San Diego County 2·11·31 Posts	It will probably be more like 2 numbers in 18.5 days. Each number can run on its own core, but one cannot expect 100% linear scaling.

2006-08-21, 20:50	#3
BlueCatZ1 Aug 2006 2×3 Posts	On a X6800 Conroe in XP MCE here are some numbers: With one instance of P95 at 2048 fft I was getting .041 sec per iteration. With one instance LL testing and one in P1 factoring I was getting .051 sec/iter. With two instances of P95 at 2048 fft I was getting .046 seconds/iteration. So 16.9 Days for One 10M digit LL test or 18.97 Days for two? Do these numbers sound right? Hope this helps. -Zak

2006-08-24, 18:07	#4
PrimeCrazzy Dec 2005 E₁₆ Posts	Those numbers sound great. I currently am running a P4 at 2.9Mh. I need to convince my wife it is time for an upgrade.

2006-08-24, 19:03	#5
Mystwalker Jul 2004 Potsdam, Germany 3·277 Posts	I would be interested in gmp-ecm benchmarks of the Core2. Does anybody care to run some curves with it?

2006-08-28, 03:21	#8
BlueCatZ1 Aug 2006 110₂ Posts	Here are results from ECM (Version I was able to compile)(Was saying stuff for pentium3 durring install?) GMP-ECM 6.1.1 [powered by GMP 4.2.1] [ECM] Input number is (300 digits) Using B1=1000000, B2=1045563762, polynomial Dickson(6), sigma=2996848946 Step 1 took 47390ms Step 2 took 18485ms (Version from forums for Prescott p4) GMP-ECM 6.1.1 [powered by GMP 4.2.1] [ECM] Input number is (300 digits) Using B1=1000000, B2=1045563762, polynomial Dickson(6), sigma=1801303844 Step 1 took 41703ms Step 2 took 17109ms Tests done on Core 2 Duo X6800 with With 4 GB Ram. Hope This helps. -Zak

2006-08-28, 19:20	#9
Mystwalker Jul 2004 Potsdam, Germany 3×277 Posts	Thanks for the testing, Zak! Even with binaries that aren't optimized for the CPU, it is 40% (Step1) to over 60% (Step2) faster than a P4 Prescott - at least for this first test. I wonder how the Core2 performs in relation to an Athlon64. I guess that currently, the A64 will pull away (due to assembly support), but the Core2 could be a real competitioner once the software is optimized.