Best way to set up systems for maximum output

[blert]

I used to casually crunch several years ago and am looking at getting back into it. I recently built a quad core system for gaming and for the most part it just sits there smugly churning away at nothing most of the day. There will be another system that will be going along for the ride.

Both system listed in signature below. I will be switching out the FX-57 for a 4000+ Clawhammer on the secondary machine in the near future as I can get a better OC on that cpu.

My goal is to have both of the rigs running as efficiently as possible in regards to productivity (e.g.- all cores folding- or not depending on current wisdom, possibly the GPU cores folding as well). Will also like the ability to shut down everything (folding wise) so that when I do game on the quad core all of the resources will be directed towards that, then easily started up again.

I have read thru many of the posts here and honestly I'm more confused than when I started to dig for info. If anyone is willing to give me some specific instructions or can point me in the right direction that has baby steps it would be much appreciated.

Thanks!

 

[Philippart]

Welcome back, we all appreciate your contribution!

There isn't a Nvidia GPU client so far, so you won't be able to let them crunch.
In order to get the most output from the computers, you should run the SMP client on the quadcore and the regular client on the 4000+, of course in console version.
Set the clients to get "big" WUs, they'll give you some nice bonus points.

Another tuning method for the SMP client on the quadcore is to run the SMP client on linux x64 in Vmware. The linux (SMP) client is far more efficient than the windows version, it's the opposite for the regular singlecore client.
If you don't like the idea of running linux virtually you can set the affinity of the instances manually to 1 core each, there's also an application that does the same, however security issues have been raised concerning it.

We're probably having a F@H race in December, I'll post a step by step guide on how to run the linux SMP client in Vmware on windows this week.

EDIT: running the SMP client in Vmware makes a difference of around 220 ppd on exactly the same Core2Duo system.

 

[blert]

Thanks for the welcome back!

So, as I'm going thru all of the information it's all starting to come back.

The question I now have for the group: on the quad core system is it better to run one instance of F@H or to run 4 separate instances? Will it process 1 WU very quickly or should it be set up to plug away at 4 WU's? Does it make any difference one way or another? Sorry for the noobish questions but it's my first multi-core system and I want to get the most out of it....

Memory isn't an issue as there is 4GB installed.

OS is WinXP Pro SP2- I know, I know... looks like I'll have to dust off the linux stuff soon and configure a dual boot system....

 

[RaySun2Be]

Welcome back to Team Anandtech, Blert!

 

[natethegreat]

Originally posted by: blert
The question I now have for the group: on the quad core system is it better to run one instance of F@H or to run 4 separate instances? Will it process 1 WU very quickly or should it be set up to plug away at 4 WU's? Does it make any difference one way or another? Sorry for the noobish questions but it's my first multi-core system and I want to get the most out of it....

1 instance of the F@H SMP client will use all 4 cores simultaneously. This just happens to be the best ppd client also, so running the Linux or Windows SMP client will be the the best use of your quad core rig. I actually run the SMP client on all 4 of my dual cores boxes even though F@H reccomends at least quad core systems for the client. I get a little over 500 ppd average over the 4 dual core rigs.

 

[GLeeM]

Welcome back, blert

The SMP client will give you the best points per day. It will run four instances of FahCore_a1.exe on one WU.

The drawback is that it is a little more involved install, if you follow instructions you should be OK.
Also, it is a late beta client and is sensitive to network connection disruptions and a few other known problems. Some computers have a problem with restarting the SMP client without losing work already done. SMP gets good points because it is a high performance client - meaning it needs a newer computer to finish before the short deadlines.

I would give it a try and if you have no problems then great and if you do have problems then try four regular clients.

 

[Uppsala9496]

The SMP client is the way to go. As GLeeM mentioned there are some network connection hangups now and then, but it's worth the minor inconveniences.

I have it running on both a quad core and a dual core.