[orion456]

I have several old PIIIs folding and I am thinking they use more energy to fold than they are worth. Are PIIIs really doing anything that can't be done more efficiently by Quads? What do you think about shutting them off?

What about P4s. Right now mine are supporting ati x1900 GPU folding, but when that ends shortly, are P4s really worth folding by themselves (energy and science wise)?

[thideras]

The PPD:Power ratio is horrible. It is not worth it.


EDIT: You said P3 in your title and meant P4 also >.>

It is up to you, I'd sell them and get Core2's

[orion456]

Originally Posted by thideras

You said P3 in your title and meant P4 also

I thought about my P4s after I made the title.

[WarriorII]

All WUs need being done.

Single Client, SMP, what ever.
They all need done.

The majority of T32 has moved on to C2D for more efficient Folding.
(even though we OC the heck out of them & make them not efficient anymore)

Intel C2D & Quad is really the way to go.
Sell your stuff while you can get a fair price for it & move to the Next Gen stuff.

[benbaked]

PIIIs are better for seti@home then folding, and then only if they are 1 GHz or faster IMO. At least in SETI they can complete a WU in around eight hours or less, with folding you're looking at a week or more per WU.

It's not worth it. At least not until Stanford increases the point values of uniprocessor client WUs, and even then I doubt it will be worthwhile.

[Evilsizer]

FAh needs to start making use of the instruction set's on the cpus like SETI/BONIC. i see a huge difference in PPD for my mobile cpu rig in sig. pegged at 1k PPD, not back for only 1.86ghz..

[ChasR]

By configuring for -advmethods without big WUs, you can get about 100 ppd/GHz out of the p3s on bonus Amber WUs, until August. The P4s should crank out about 70 - 80 ppd/Ghz on the Ambers. I'm still doing it, but not enthusiastically.

[Zerix01]

Has anyone even tried clustering four P4's and running the SMP client. I know people were saying that the network connections would be too slow, but has anyone even tried this? This could give more life to those boxes. Not terribly great on power but hey it would be more interesting.

[APOLLON]

Tonight, Stanford officially shut down the GPU1 project and my 3.4GHz Northwood P4 is folding on the classic console client. My PPD went from ~600PPD down to about 150PPD, and that's with the bonus amber WUs. My dual PIII 1.1GHz has two console clients running and together they produce anywhere from 160-200PPD. How can an older system with a combined frequency of 2.2GHz beat a more recent system @3.4GHz? Is Netburst that bad? I'm honestly thinking of keeping my dual PIII and shutting down the P4 or get an AGP 3850 for it.

Originally Posted by Zerix01

Has anyone even tried clustering four P4's and running the SMP client. I know people were saying that the network connections would be too slow, but has anyone even tried this? This could give more life to those boxes. Not terribly great on power but hey it would be more interesting.

Unfortunately, no client supports clustering. Apparently, Stanford is looking into it but no one has stepped forward to lend them the support they need to move forward with the project. They are researching it, this much I know.

[thideras]

Originally Posted by APOLLON

Is Netburst that bad?

Yes, very much so.

[JDXNC]

I wouldn't fold on my PIII's if they weren't already running 24/7 in my server, but seeing as they are, they might as well fold too.

[Zerix01]

Originally Posted by APOLLON

Unfortunately, no client supports clustering. Apparently, Stanford is looking into it but no one has stepped forward to lend them the support they need to move forward with the project. They are researching it, this much I know.

The SMP client will run on any system reporting 2 or more processors. Generally when clustering in Linux the system will look like an SMP system with how ever many processors in one box, but physically they are completely separate systems. So yes they don't support clusters but the client should be able to be fooled in to running on one simply because a multi core system and a cluster should not look any different from the kernel side.


Ah one thing just thought of though, those P4's would need to support x86-64 instructions.

[ChasR]

IIRC you'd need a 10 Gb/sec ethernet connection to make SMP work efficiently on a cluster. THe investment would be better spent on a C2Q.
And yes, you'd have to have 64 bit capability.

[Evilsizer]

chasr,
im not dougting you but i figured the size of the cluster would determine the bandwidth needed on the network. simply cause no matter the speed its travling as fast as it can over copper/fiber lines on a network. there would be a bigger latency hit with 4 boxes vs a since quad cpu box. was any info about clustering ever posted on FAH site? i did look a while back and never found anything.