Building DC farms; Experience, advice, questions and answers - Page 2

Nebob · 2002-09-06, 02:54

The Asus P4S533 has integrated LAN and netboot support. No bios hacking required :)

Lord_Humungous · 2002-09-09, 17:14

What HSF is the prime monster running? OEM? I looked over the Monster site quickly and didn't see a HSF model listed.

I saw in another thread your temps are running 10-12C over system temp and I was curious.....

Prime Monster · 2002-09-09, 19:53

Quote:

What HSF is the prime monster running? OEM? I looked over the Monster site quickly and didn't see a HSF model listed.

The default HSF.....

ET_ · 2002-09-10, 12:41

I'm planning to build a small monster_farm.

I can buy a system based on Athlon XP 2000+ OR Pentium4 1.5G at approx. the same price. I'm aware of the better performances of Pentium 4 with George's code (Prime95, PRP, ECM), but my goal is to join different mathematical DC projects, not only GIMPS.

The question is: How faster is Athlon vs. Pentium 4 with "normal" code (i.e. non SSE2 assembly optimized) provided they have the same frquency clock?

Luigi

Tasuke · 2002-09-10, 20:33

The athlon tends to be about twice as fast in non-bandwidth sensitive apps, 50% faster when bandwidth is an issue. Gimps the P4 is 50% faster clock for clock. in rc5 and eccp, you get worse than the 2X due to the lack of a barrel shifter. P4 is not a crypto processor.

crash893 · 2002-09-16, 17:24

my other board came up with this

mini diskless farm

http://forum.oc-forums.com/vb/showthread.php?s=&threadid=122568
enjoy

Mivacca2 · 2002-09-17, 23:17

I am thinking about using a computer lab (with permission of course since i work here) and every so often we have to image all the machines which entails making them all exactly the same by doing a scratch install and just copying a HDD over and over again to different machines and going in and changing the names... and i was wondering if gimps would pick up on this since i would have at one time 24 different exponents out and then only have 24 computers working on one, would that be like catastrophic on either end. That is really the only thing that I am worried about. If that is completely incoherent then I will try to rephrase the question. THX for the help

MiVacca2

garo · 2002-09-18, 01:08

Yes it won't work as you describe. A good alternative if all these machines have shared directories is to have all the machines work off directories in a central server. Greeners posted a question about this in another thread. Do a search.

binarydigits · 2002-09-18, 01:30

Won't they still have the same problem after imaging? That is: if they are set up to use 24 different directories on a central server, then after imaging won't they all end up using the same directory on the server and ignoring the other 23? (I am assuming that when the HDD is duplicated it copies everything including the registry.)

Prime95 · 2002-09-18, 03:22

As to your question, qill the server be able to cope? I believe the answer is yes.

As to best implementation, you can try make sure worktodo.ini is empty prior to the mass copying of the HDD. I'd choose your prefered work type based on how often this copying happens. The more often it occurs the more more I'd push for shorter work units.

The only downside is you will see lots of exponents reserved that will have to expire after 60 days inactivity.

Prime Monster · 2002-09-18, 06:49

You can do a couple of things:
[list]Use a small script when starting prime the first time after an install to insert the machine name in the local.ini file. You could probably just append it[/list:u][list]Set up a directory structure on your server with subdirectories named after your machine names, link that directory at boot, start prime from that directy. This is technically what I do for the Prime Monster[/list:u]

Alf

2002-09-10, 12:41	#15
ET_ Banned "Luigi" Aug 2002 Team Italia 4818₁₀ Posts	Pentium4 vs. Athlon XP I'm planning to build a small monster_farm. I can buy a system based on Athlon XP 2000+ OR Pentium4 1.5G at approx. the same price. I'm aware of the better performances of Pentium 4 with George's code (Prime95, PRP, ECM), but my goal is to join different mathematical DC projects, not only GIMPS. The question is: How faster is Athlon vs. Pentium 4 with "normal" code (i.e. non SSE2 assembly optimized) provided they have the same frquency clock? Luigi

2002-09-17, 23:17	#18
Mivacca2 Aug 2002 2·5 Posts	Quick question I am thinking about using a computer lab (with permission of course since i work here) and every so often we have to image all the machines which entails making them all exactly the same by doing a scratch install and just copying a HDD over and over again to different machines and going in and changing the names... and i was wondering if gimps would pick up on this since i would have at one time 24 different exponents out and then only have 24 computers working on one, would that be like catastrophic on either end. That is really the only thing that I am worried about. If that is completely incoherent then I will try to rephrase the question. THX for the help MiVacca2

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Any answers on missing factors?	schickel	Aliquot Sequences	8	2011-11-29 12:24
How do you track your farms status?	mathemajikian	PrimeNet	2	2007-07-24 21:10
Need advice for building a new computer	garo	Hardware	20	2006-12-31 22:50
GGNFS questions and Answers	VJS	Factoring	100	2006-01-15 08:34
Monsters and Monster farms	Unregistered	Data	6	2004-08-12 00:28

2002-09-06, 02:54	#12
Nebob Sep 2002 11 Posts	The Asus P4S533 has integrated LAN and netboot support. No bios hacking required :)

2002-09-09, 17:14	#13
Lord_Humungous Aug 2002 23 Posts	What HSF is the prime monster running? OEM? I looked over the Monster site quickly and didn't see a HSF model listed. I saw in another thread your temps are running 10-12C over system temp and I was curious.....

2002-09-10, 20:33	#16
Tasuke Aug 2002 2³×3² Posts	The athlon tends to be about twice as fast in non-bandwidth sensitive apps, 50% faster when bandwidth is an issue. Gimps the P4 is 50% faster clock for clock. in rc5 and eccp, you get worse than the 2X due to the lack of a barrel shifter. P4 is not a crypto processor.

2002-09-16, 17:24	#17
crash893 Sep 2002 2³×37 Posts	my other board came up with this mini diskless farm http://forum.oc-forums.com/vb/showthread.php?s=&threadid=122568 enjoy

2002-09-18, 01:08	#19
garo Aug 2002 Termonfeckin, IE 2²·691 Posts	Yes it won't work as you describe. A good alternative if all these machines have shared directories is to have all the machines work off directories in a central server. Greeners posted a question about this in another thread. Do a search.

2002-09-18, 01:30	#20
binarydigits Aug 2002 2²×13 Posts	Won't they still have the same problem after imaging? That is: if they are set up to use 24 different directories on a central server, then after imaging won't they all end up using the same directory on the server and ignoring the other 23? (I am assuming that when the HDD is duplicated it copies everything including the registry.)

2002-09-18, 03:22	#21
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 1110101110001₂ Posts	As to your question, qill the server be able to cope? I believe the answer is yes. As to best implementation, you can try make sure worktodo.ini is empty prior to the mass copying of the HDD. I'd choose your prefered work type based on how often this copying happens. The more often it occurs the more more I'd push for shorter work units. The only downside is you will see lots of exponents reserved that will have to expire after 60 days inactivity.

2002-09-18, 06:49	#22
Prime Monster Aug 2002 404₈ Posts	You can do a couple of things: [list]Use a small script when starting prime the first time after an install to insert the machine name in the local.ini file. You could probably just append it[/list:u][list]Set up a directory structure on your server with subdirectories named after your machine names, link that directory at boot, start prime from that directy. This is technically what I do for the Prime Monster[/list:u] Alf