Most economical blade style farm?

As the thread title suggests. I realize that doing it has probably fell off a cliff recently with the advent of -bigadv, but was wanting some input on it anyways.

With the volatility of Stanford's decisions surrounding -bigadv I will not risk investing in a portion of the project just to be kicked out of it by another "project restructuring". So my interests are purely SMP/GPU units and what hardware is the best bang for the buck to that end.

My main concern is PPD/KWH. I don't want to put together anything that will be pulling 800 watts at the socket either.

Also when I say economical, I guess my rough number would be at most $600 per complete individual blade. If that's just unrealistically low, let me know, but I think it should be doable.

Here is what I have as a prototype just dicking around on newegg:

Motherboard: Asus P8H61-M http://www.newegg.com/Product/Product.aspx?Item=N82E16813131793

CPU: Intel Core i7 2600K http://www.newegg.com/Product/Product.aspx?Item=N82E16819115070

RAM: Corsair 2x2GB 1333 DDR3 http://www.newegg.com/Product/Product.aspx?Item=N82E16820145278

Cooler: Cooler Master Hyper 212 http://www.newegg.com/Product/Product.aspx?Item=N82E16835103065

PSU: Coolmax 600 Watt junky piece o' crap http://www.newegg.com/Product/Product.aspx?Item=N82E16817159055

HDD: WD Caviar 40GB recertified http://www.newegg.com/Product/Product.aspx?Item=N82E16822136897

Totals comes to $518.81 shipped.

~FA

You've got some nice choices in there. Very sweet. Video?

The way I see it for cpu folding is, that they will need bigadv going forward. They've done a LOT of small protein simulations. Now the opportunity is there because of the hardware, to study the bigger proteins, and every research group is anxious to have something studied - Pande Group will have no problem coming up with candidates to be studied.

Most server boards are 2P. The type that PG wants to use are the 6174 class, or better. With a 4P board, even with 6272 cpu's (which are slower), Tanker (my 4P folder), is returning bigadv with 75% of the time still remaining.

That means it will be at least 5 years before Tanker becomes unable to complete a bigadv work unit, within the deadline. After that, Tanker can still fold regular SMP work units, very quickly, and make good ppd.

I expect Tanker to be a worthwhile folder for 10 years, with one upgrade of the cpu's, after 5 years. The G34 socket has already been given another generation of updates, on the future track for AMD.

I don't see a future for any Intel socket, going forward that far. Not a problem for a gamer, because they'll want to upgrade by then, anyway. But for a folding farm, it's a real problem, imo.

If the bigadv program were eliminated (and I don't believe that will happen, it's here to stay), Tanker would still be (and I'm guessing here), about equal to 4 2600K Blade systems, unless you overclock them. And if you do overclock them, then the cost for cooling and wattage will dramatically increase.

Say you got 25k ppd from that little darling. How would it match up with one 4P system:
*This is a little low, but I believe the future points for bigadv will be decreasing a bit.

**At the wall readings with a Kill-A-Watt meter, for the entire system

The Blades are very tempting, but having a look at the figures, you see the efficiency you get from one system, instead of several.

Regardless of what happens or doesn't with BA work, the QRB will more generously reward more cores in a single machine as opposed to the same number of cores in different machines. I saw a result published on FF.org wherein a donor ran one WU on each of his 6 core processors in his 2P machine, then compared the result to running a single WU run on both processors. THe single WU produced 50% more ppd than 2 WUs. That is the nature of the QRB and that's not going away soon.

I guess I'm just worried with the track record of bigadv that they will further drop deadlines out of range of the run of the mill opteron server.

Adak, do you have any numbers on how the opterons in 4p systems fold standard SMP units? PPD etc..?

Is a 50% bonus for running one 2P vs two 1P machines not enough? BA has nothing to do with that.

I guess I wasn't clear on that point, thanks.

In reality, it's more complex. You have to pay a lot for xeon proccessors to go in an intel based 2P rig. AMD 2P rigs are lacking in performance compared to intel 2P, but are cheaper. IIRC, the 4P 6174s make about 220,000 ppd on regular smp work, about the same as 10 2600Ks, comparing both at stock clocks. Reduce that to 6 2600Ks at 4.8 GHz.

@FoldingAddict
No hard data, I just recall reeling off my first ten regular SMP wu's in Tanker - it was fun just watching the finished % lines of text, roll up the console window.

I could give you some exact figures after the CC, if you like.