Converting from v4 to v5

[devjonfos]

I'm running Fedora 9 and I converted from Prime95 (mprime) v4 to v5. I followed the upgrade instructions on the home page (http://www.mersenne.org/freesoft/) but got bogged down in several places.

My system:
==========================
XPS 630 Q6600 @ 3.20GHz OC
Windows XP Pro/Fedora Linux
4GB DC DDR2 @ 800MHz
2-GeForce 8800 GTs 512MB
2-500GB / 1-160GB ALL SATA
16X CD/DVD Burner Single Drive
Sound Integrated 7.1 Channel Audio

So I have four cores in my Q6600 which is OC'd to 3.20 GHz.

First problem was "libcurl.so.3" was not found. So I did a yum install of the libcurl package. Same error. So I did a search and found "libcurl.so.4". So I made a soft link from "libcurl.so.3" to "libcurl.so.4" and that seemed to work.

Second problem was understanding the interface presented when doing an "./mprime -m". I didn't understand it very well and I went through the options several times. Finally, I got most of it correct except how to get mprime to run on each of my cores.

Third and final problem was to get mprime to work on each core of my quad-core.

I ended up editing the "local.txt" file and adding Affinity lines to each Worker Thread.

WorkerThreads=4
ThreadsPerTest=1

[Worker #1]
Affinity=0

[Worker #2]
Affinity=1

[Worker #3]
Affinity=2

[Worker #4]
Affinity=3

And my "worktodo.txt" looks like this:

[Worker #1]
Test=529F1EEC5480B4717F778991D6513D02,45706883,68,1

[Worker #2]
Test=BB6BEE1AAB8135D0A508D8C1738FAFF4,46232089,68,1

[Worker #3]
Test=24D58441807842D7CF220505D2BAECE6,46232201,68,1

[Worker #4]
Test=521CEDDA57130B43C3654395E3E808F9,46232213,68,1

So far this is what I get from "./mprime -d":

[Main thread Jan 27 21:21] Mersenne number primality test program version 25.8
[Main thread Jan 27 21:21] Starting workers.
[Worker #2 Jan 27 21:21] Worker starting
[Worker #2 Jan 27 21:21] Setting affinity to run worker on logical CPU #1
[Worker #4 Jan 27 21:21] Worker starting
[Worker #4 Jan 27 21:21] Setting affinity to run worker on logical CPU #3
[Worker #1 Jan 27 21:21] Worker starting
[Worker #1 Jan 27 21:21] Setting affinity to run worker on logical CPU #0
[Worker #3 Jan 27 21:21] Worker starting
[Worker #3 Jan 27 21:21] Setting affinity to run worker on logical CPU #2
[Worker #4 Jan 27 21:21] Starting primality test of M46232213 using FFT length 2560K
[Worker #2 Jan 27 21:21] Resuming primality test of M46232089 using FFT length 2560K
[Worker #3 Jan 27 21:21] Resuming primality test of M46232201 using FFT length 2560K
[Worker #1 Jan 27 21:21] Resuming primality test of M45706883 using FFT length 2560K
[Worker #3 Jan 27 21:21] Iteration: 58 / 46232201 [0.00%].
[Worker #2 Jan 27 21:21] Iteration: 50 / 46232089 [0.00%].
[Worker #1 Jan 27 21:21] Iteration: 2061514 / 45706883 [4.51%].
[Worker #1 Jan 27 21:44] Iteration: 2070000 / 45706883 [4.52%].  Per iteration time: 0.166 sec.
[Worker #2 Jan 27 21:46] Iteration: 10000 / 46232089 [0.02%].  Per iteration time: 0.152 sec.
[Worker #4 Jan 27 21:47] Iteration: 10000 / 46232213 [0.02%].  Per iteration time: 0.154 sec.
[Worker #3 Jan 27 21:47] Iteration: 10000 / 46232201 [0.02%].  Per iteration time: 0.158 sec.
[Worker #2 Jan 27 22:12] Iteration: 20000 / 46232089 [0.04%].  Per iteration time: 0.153 sec.
[Worker #1 Jan 27 22:12] Iteration: 2080000 / 45706883 [4.55%].  Per iteration time: 0.164 sec.
[Worker #4 Jan 27 22:12] Iteration: 20000 / 46232213 [0.04%].  Per iteration time: 0.154 sec.
[Worker #3 Jan 27 22:13] Iteration: 20000 / 46232201 [0.04%].  Per iteration time: 0.156 sec.

And all four cores are pegged, which is what I would expect, and which is what the prior behavior was in v4 using the -A switch.

This isn't a rant, I just want to present what I went through, try to provide info to others, try to provide info to developers, and get some feedback as to whether I got it right, or what I need to change.

[petrw1]

I can't speak for Linux as I have never used it; however your per-iteration times seem very high. If you check out the benchmarks page you will see that most others with a Q6600 @ 3.2 are getting under 0.05 seconds per iterations for exponents in your range. Even allowing for contention of all 4 cores I would expect it to still run under 0.06.

http://www.mersenne.org/report_benchmarks/

[S485122]

Originally Posted by petrw1

I can't speak for Linux as I have never used it; however your per-iteration times seem very high. If you check out the benchmarks page you will see that most others with a Q6600 @ 3.2 are getting under 0.05 seconds per iterations for exponents in your range. Even allowing for contention of all 4 cores I would expect it to still run under 0.06.

http://www.mersenne.org/report_benchmarks/

I will bet the motherboard has a NVidia chipset. That kind of chipset are good for a dual core at the most, then memory contention quicks in badly with Prime95 or mprime.

Another factor is that the overclock is surely achieved via increasing the FSB clock from 1066 to 1422 with 800MHz memory...

That said I can offer no help on mprime. :-(

Jacob

[devjonfos]

Originally Posted by S485122

I will bet the motherboard has a NVidia chipset. That kind of chipset are good for a dual core at the most, then memory contention quicks in badly with Prime95 or mprime.

Another factor is that the overclock is surely achieved via increasing the FSB clock from 1066 to 1422 with 800MHz memory...

That said I can offer no help on mprime. :-(

Jacob

Correct on both points (NVIDIA chipset and OC via FSB increase to 1422).

[garo]

I would recommend changing the work on one core to TF. Stop prime.txt, change the work preference for worker 4 using ./mprime -m. Then move the LL test line unser worker #4 to worker #3. And add the line UnReserveDays=90 in your prime.txt to make sure you don't accidentally unreserve the second LL test on worker 3. Restart and watch the lower iteration times.