Well, my interesting note is that Vista is more stable than Linux. Specifically, the C2Q9450 on an MSI X48c Platinum with 2GB OCZ low latency DDR3 would lock up frequently - at least daily - when running the SMP Folding client on Ubuntu. So far it has been rock-solid stable on Vista Ultimate with 4 single-core Folding instances running. In fact, under Linux I couldn't OC the rig at all and even had to slow the LL RAM to more normal speeds. Under Vista I'm now slightly OCing the machine.
I've actually just ordered Windows Home Server & the 1.5TB Seagate and will re-do the machine as my AV server & workstation backup platform. I'll eventually add dual 4870s if the price comes down for extra Folding goodness.
The rest of my farm is pretty pedestrian: 4 Q6600s (3xVista, 1xXP; 1x4GB, 3x2GB, 3x HD3870; 3x500GB Samsung, 1x80GB Fujitsu notebook drive) with no serious overclocking, 1 C2D T7800, and a lowly 2GHz P4M. Not all are on 24x7 and during the hieght of summer I powered most of it down to cut the electric bill. A lot of it is in my basement where the heat & noise aren't an issue; with cold weather approaching I may distribute them around the house in rooms we don't use much to take advantage of the generated heat. Partially offset the electricity with reduced natural gas consumption.
I'm sure I could have higher PPD if I tried and did some OCing or other workarounds, but it is fine for the moment. It's easier to throw a few $ at it than find the time & desire to tinker.
As winter approaches, I'm seriously considering getting some Nvidia cards (9800X2 or 9600 Gso) to GPU fold as more efficient than powering up the X2 machines that I shut down during the summer (They were powered up as needed, rather than 24x7 folding). I just don't think that $60 and 80W per 5000PPD (9600 Gso) or $260 (9800X2) and 180W per 20,000PPD is that bad of a price compared to $0 and 220W for 500PPD (SMP on a 4600+ X2).
I just feel that it is just better to power up fewer machines with GPU's and get my needed heating wattage via GPU's than CPU's and thereby fold for a lot more points.
The risk is that Stanford fixes the ATI GPU client and then the Nvidia clients PPD will drop like a rock. At the moment all the WU's are benchmarked against an ATI 3850 for 1500PPD. Because the ATI clients suck big ones compared to Nvidia, Nvidia gets high PPD. That all changes if they fix the ATI client so that the productivity of the ATI cards increase comparable to their capabilities. Then the Nvidia won't have the massive computational advantage and their PPD's will drop like a rock while ATI stays stable because everything is based on 3850=1500PPD
I have added two folding video cards to machine #1 and reorganized the HW somewhat over the time period. So here is my current modified configurations.
Intel Q9450 (OC to 430x8=3.44GHz); 4x2GB=8GB Ram; Gigabyte GA-X48-DS4 Motherboard; PC Power and Cooling Q750 PS; 2x Asus 9600GSO; Windows Server 2008 X64 (SP1).
I am folding using two instances of Ubuntu 8.04 under Hyper-V beta (M$ virtual machine) each one has two proccessors and 2GB RAM and is running the SMP client producing roughly 3000PPD (p2662) each. I am also running two instances of the NVidia GPU2 client on my 9600GSO's each producing around 3000PPD. The machine uses 333W while folding and produces around 10,000-12,000.
Intel Q6600 (OC to 400x8=3.20GHz); 2x1GB=2GB Ram; Gigabyte GA-X48-DS4 Motherboard; PC Power and Cooling Q750 PS; 2x Sapphire 3870 video card; Windows XP Media Center Edition (SP3).
Here I am folding two clients the GPU2 and the SMP. I am getting 2000PPD each for the GPU2 clients (p4731) and 1900PPD for the SMP client (p2665) for a total of 5900PPD. The machine is using 307W while folding.
So in theory, I should be generating around 16,000-18,000 PPD with the two machines which is significantly higher (around 3x PPD, while using less than 1/3 of the power) than I was getting with the previous 9 machines.
I have found that to consistantly keep A2 cores in Linux (by far the most productive core) one needs at least 2GB of RAM per machine. Perhaps your Linux users can use that info to get 2x the PPD of the A1 cores.
I have found that having multiple video cards folding definately produces user-interface lag that is quite annoying for normal use. Multiple video card folding is better done on dedicated folding machines.
Linux is more productive than Windows; Nvidia is more productive than ATI; and an OC'ed Q9450 is better than an OC'ed q6600.
I have no idea. I haven't seen any data on it yet. My assumption is that the i7 should do very well, though quite spendy and it still won't compete well against GPU folding.
You can get MB's for the C2D with upto 4 PCI-E slots. There is no significant PPD lost till 4x and even then it isn't bad. The big problem with running 3-4 video cards is heat, power, and the required monitor connections (building null-monitor plugs so that the drivers think the card is active), but not the CPU. I've read that there are several people running three 9800GX2's with six clients on a single MB and that's 30,000PPD just from the GPU folding. From the discussions it was a bear to get running properly though.
If Stanford was properly scaling the computational ability of video cards then that number would be much higher. But they don't want to diminish and alienate all their traditional folders. GPU's really can out calculate a traditional CPU by a very long ways.
Stanford is actually collecting statistics on how much calculating is being done by the different clients (These are real numbers, not estimated peak FLOPS): F@H client stats. The entire productivity of all convential CPU's currently folding (Windows, Linux, Mac) could be replaced by just a little over 2,000 folding GPU's.
I thought I would show you what a determined folder can do: 63,ooo PPD using 1124W from a single machine. The only issue, I see, is how reliable considering pushing that single PS and the total heat load of the machine. I suppose that time will tell. I wonder if he ever seals the case.
It is my understanding that he intends to create a folding farm contining multiple of these all in a rack.
The machine specs:
Motherboard MSI K9A2 Platinum @ $145
AMD 9950 CPU @ $163
Memory 2GB DDR2 @ $22
Thermaltake PSU (1200w) @ $309
4 x GTX295 @ $505 = $2020
250GB HDD @ $40
CPU Cooler @ $40
Windows Vista @ 190