Folding on Opterons

[Horatio]

Hey everyone, I haven't been able to find much information about folding on Opteron processors. I saw this thread,

http://www.hardforum.com/showthread.php?t=1398636

And it got me thinking about folding on those bad boys, seems pretty damn cheap for 8 Cores power! However I can't seem to find ANY information about folding on an opteron, what kind of PPD etc. Is there a reason nobody is building with these? (Other then the TLB bug)?

 

[Zero82z]

PPD is lower than comparably-clocked Core 2 CPUs, and these run hotter and consume more power. GPUs also put out far more points, cost less overall (you don't need a server board, RAM, etc.), and use less power.

Basically, at best you'd eke out about 3k PPD per chip (with LinSMP and a2 workunits), and you'd have to spend money on motherboards, RAM, PSUs, etc. A single GTX260 can be found for around $200 or even less, and will put out more points than two of these quad-cores combined. So this really isn't a particularly good route to go for a dedicated F@H box.

 

[osrk]

I've folded on those CPU's it's not worth it. Like zero said just the cost of the CPU and you could buy a video card and fold for 2x the speed.

 

[EvilAlchemist]

I really feel we are becoming a GPU top heavy Team ....

It used to be that we would help anyone out ad encourage them ...

Now all we say is --> GPU .. Damn the Points ... let just help people get to folding on SOMETHING

Some of you are going to shoot your self in the foot if something happens to PPD or the Client

I am getting some good PPD on my CPU Clients.
Q9650 @ 2.72 gets approx 5500 PPD with Linux SMP --> Pulling 125 Watts
(That is with a 8 x Raid 5 Array steaming video to 2 other systems on the network)

My Laptop is getting 2800 PPD with Linux SMP T9300 @ 2.5ghz
That is damn good for that system.

---------------------------------------------------------------------------

Back to the OP's Question

I would suggest looking at this thread --> http://foldingforum.org/viewtopic.php?f=38&t=7053

4x Opteron 8347 getting around 11,000 - 12,000 PPD ...

I think he has replaced all them with HE models to lower power usage.

It would be a fun project to get up and running

 

[osrk]

I really feel we are becoming a GPU top heavy Team ....

It used to be that we would help anyone out ad encourage them ...

Now all we say is --> GPU .. Damn the Points ... let just help people get to folding on SOMETHING

Some of you are going to shoot your self in the foot if something happens to PPD or the Client

I am getting some good PPD on my CPU Clients.
Q9650 @ 2.72 gets approx 5500 PPD with Linux SMP --> Pulling 125 Watts
(That is with a 8 x Raid 5 Array steaming video to 2 other systems on the network)

My Laptop is getting 2800 PPD with Linux SMP T9300 @ 2.5ghz
That is damn good for that system.

---------------------------------------------------------------------------

Back to the OP's Question

I would suggest looking at this thread --> http://foldingforum.org/viewtopic.php?f=38&t=7053

4x Opteron 8347 getting around 11,000 - 12,000 PPD ...

I think he has replaced all them with HE models to lower power usage.

It would be a fun project to get up and running

It's not that the OP has the processor it's that he's wondering if they'll fold well. The answer is no. A $125 video card will beat those things. Trust me I had a dual 8346HE rig last month, the processors did not fold well at all. It was basically overkill. The Pande group is pushing people towards GPU's anyways with the point assignments.

I don't care about the points. Damn the points. I use them as a performance indicator i.e the big one I pay attention is ppd/W, those cpu's have a horrible ppd/W. If the OP has the money for dual quad's with all the subcomponents i'd rather suggest he make a dual GPU box. Not only would it be cheaper it would also scale very well in the future.

 

[lelliott731]

I agree with Evil, GPU's can only do a certain type of WU, whereas CPU's can do every type of WU. I think they are even using regular CPU's to double check GPU work & even SMP work. I forgot where I read that, but I'm pretty sure I did.

Either way, we don't want to only be doing this for points, we want the most help for science; this is a great project that could help a lot of people who are suffering from a lot of horrible illnesses.

 

[Wesleynator]

We are all right here. Science is important obviosly. If it weren't, I don't think there would be so much passion about our purpose. Points are important too because it fosters competition and increased participation. I run about every client out there (GPU, SMP, Classic, PS3) because they all help and they all make points and also because it's what runs best on the machines I have. If my electric bill gets too high this summer, I'll probably shut down the PS3 because it is not nearly as efficient from a ppd/w standpoint. No matter what clients a fellow team member chooses to run, (all GPU, all SMP or a mix) - it's all good. I don't think anyone is saying don't run anything but GPU but if someone asks about the relative merits of one client vs another taking into account points, watts, etc., then it's clear Stanford has made the GPU very attractive.

 

[Zero82z]

I really feel we are becoming a GPU top heavy Team ....

It used to be that we would help anyone out ad encourage them ...

Now all we say is --> GPU .. Damn the Points ... let just help people get to folding on SOMETHING

Some of you are going to shoot your self in the foot if something happens to PPD or the Client

Well, the thing is that it isn't cost-effective or energy-efficient to put together an 8-core or 16-core Opteron box exclusively for F@H. And if money is not an object, then it would make more sense to put together a box with Xeons or Core 2 Quads instead, since they use less power and are faster than Opterons. So any way you look at it, IMO an Opteron box is not a very efficient way to go.

 

[Horatio]

Thanks for the feedback guys, guess it would be more financially feasible to put together a GPU farm. Not that I'm really in a position to be able to, was more just curious why richer people haven't.