Dedicated Folding@Home Build - Advice?

[dreamscape]

Hello all

I've just upgraded my main PC's GPU to a GTX 660 so I have a spare GTX 650. I'm in the process of building a little server room and I thought why not put the 650 to use!

So I've purchased an extremely cheap desktop (HP DX2300) with a Pentium Dual Core 1.8Ghz, 1GB RAM, 160GB HDD and of course the all important PCI-E slot.

Now, here's my question. The CPU use's 65W when maxed out. Usually when I run the 650 on my i3 PC one of the cores is maxed out at 100% I guess to process work for the GPU??? Anyway. Electric prices here are pretty high so I want to try and save as much as i can without harming the GPU performance. I've noticed there are Celerons between 1.6ghz - 2.0ghz which use only 35W and have the same FSB speed as my Pentium dual (800mhz).

Could i get away with putting one of those into this machine without effecting GPU performance? I hear the 650 does just over 10k PPD normally so I want to keep it at that sort of rate.

Will a Celeron be ok for this? Also for anyone who may know.. with only 1GB RAM which would be the best OS to use with this machine? I am Linux and Windows friendly so anything goes really? Whatever will give the best performance with the very limited RAM that i have.

Thanks in advance for any help.

[Napoleon]

Windows, if you want only non-QRB CUDA core_15 GPU WUs with client v7 and max-packet-size=small setting. Those might actually produce more PPD on a 650 than the newer OpenCL core_17/core_18 WUs, since a 650 might not benefit all that much from QRB. CPU performance/power consumption won't be an issue either, since core_15 requires very little from the CPU. BTW, you *can* get core_15 WUs for Linux, but you'd need v6 GPU console client with WINE, not to mention that the configuration isn't trivial. In any case, Instructions can be found on the forum if you're interested.

Go for Linux and v7 client If newer cores (17/18) give you better PPD. There's no native core_15 for Linux, so you'd conveniently avoid getting any core_15 work. Newer GPU cores consume 100% of a single logical CPU because they simply poll the GPU to see if it's ready for new work. A lower wattage Celeron will do that "are we there yet?" nagging just as well, I think. The slower CPU will need a bit more time for setup when a WU is started/resumed and checkpointing may take slightly longer, but I think the difference in PPD isn't much for a 650.

So, I'd start with Windows and v7 to determine which type of work produces more PPD, and switch to Linux v7 if newer WUs happen to produce more.

IIRC, 1GB is miminum requirement for 32bit Win10 Tech Preview, so I guess it's (barely) sufficient for a dedicated Windows folding setup. FWIW, I'm able fold with Chrome/NaCl on a 1.5GHz Pentium M laptop with 0.75GB of memory, running Xubuntu.