50 dual-core 2.8 GHz computers: Most effective configs?

[7im]

Bruce, I wouldn't feel comfortable to ever make an odd number recomendation for SMP.

iceman1992, that's a very good recommendation. Sounds familiar...

SMP v6 Install Guide, Requirement #1

1. The system must have a recent dual core processor or better to run this client with the -smp switch in SMP mode. Four cores or more are strongly recommended. This means older single core processors cannot run this client with the -smp switch, and older dual cores will not meet the very short work unit deadlines. If after running this client with the -smp switch on an older dual or an older quad core system, and the client does not meet the work unit Timeout (formerly Preferred Deadline), please remove the SMP client and replace it with one or more instances of the CPU client and/or GPU client.

[bruce]

AMD has sold quite a number of X3's and I have not seen any reports of problems with them. Problems with 7 are pretty rare. I don't think anybody ever reported on 5, either pro or con. I've asked several times what constitutes a "large" prime and never gotten an answer, so I'm comfortable with 3 unless somebody documents a problem with it.

the words for V6 are fine with me. It never mentions 3 either way.

[Joe_H]

I recall reading a small number of reports of persons seeing a WU fail with SMP set to 5 that worked with an even number. One report was in connection with someone seeing an issue with SMP:10 set on a hex-core system to leave a core available for other processing. But reports of 5 or its multiples causing failures are rare. In all the time I have been reading this forum I have read only one mention of a failure connected to a setting of 3, and no details (i.e. documentation) were provided.

Personally I did see a couple failures with a setting of 7. But that was over a year ago so I don't recall which project(s) were involved. Since I was folding with the OS X V6 client at the time, they would have had to be core A3 ones in the 60nn-6080 range as those were the only ones available to process. But most of the rest I got processed fine with 7. Now I leave the machines at either 6 or 8, some of the projects the V7 client gets for me definitely fail with 7.

[iceman1992]

iceman1992, that's a very good recommendation. Sounds familiar...

Tried to keep it near the original. I remember reading something with wording similar to that.

[Nantes]

The dual cores appear to be handling the SMP WUs just fine. From my (limited) taking notice, when those WUs have a preferred deadline of say 5 days, the ETA is usually under a day, and when the preferred deadline is 20 days the ETA is around 3 days.

[iceman1992]

Awesome. No problems then?

[Nantes]

Well, as far as I can notice, no problems. I have only been allowed to test FAH on two machines so far. The IT director here is (rightfully) crazy about security, he means to run FAH on an isolated environment and analyze the packets it sends and etc when he has time, before he gives me full authorization to install on the other machines.

[bruce]

Well, as far as I can notice, no problems. I have only been allowed to test FAH on two machines so far. The IT director here is (rightfully) crazy about security, he means to run FAH on an isolated environment and analyze the packets it sends and etc when he has time, before he gives me full authorization to install on the other machines.

If he finds anything worth noting, please have him let us know about it.

If he finds a vulnerability that might be exploited by someone other than Stanford University (which I doubt), it's probably best to report it directly to Dr Vijay Pande, the director of FAH than making it public, especially if he knows how to fix it.

[TonyStewart14]

SMP:2 is not a problem. I've been running it on a Pentium D for some time with no trouble. The WU I have now takes 2.5 days running 24/7 and has a 12 day preferred deadline, so as long as one runs it for at least 5 hours or so per day it should be fine.