I think it's a complaint about the code quality. And yeah, I hear you, the code they use seems pretty inefficient. Although I don't agree that threads are the problem, they actually don't use threads, they use MPI to set up multiple processes with inter-process communication. Going with MPI when the windows support was dubious was a pretty bad call. Multithreading would have been a much smarter way to go from the beginning, that would have given the two platforms parity. To be fair to the Pande group, though, it sounds like they're running the simulation as a mesh diffusion, and that usually requires more communication in the form of node updates, and that may be why they went with MPI. Now sure, you can set up a multithreaded updater for a shared memory mesh without a lot of trouble, but it's a little more tricky. Hell, I could probably rewrite the stuff in python (using some C++ numeric libraries for the back end) and get better throughput, and it would be more platform independent. Although the GPU clients would be a little more tricky. But colm, they do seem to have a plan in place to improve things:
So the next version will use threads, and it sounds like they have algorithmic improvements on the way too. Now whether they'll have it ready in 'a month or two' is questionable, but it's something to look forward to.
Of course the bigger problem you may be having is that they do seem to keep renormalizing the point system for newer processors, so the P4 now seems like it barely contributes, even though previously it used to make more points. I suspect that they're getting many more calculations per point than they used to. But, the more calcs they get, the more science they can do, so it's better in that way. Eh, I dunno, I've always been a computer geek, and upgrading and setting them up is a lot of fun for me, so I enjoy it, but if it's too frustrating for you, don't drive yourself crazy.
Although it's funny, a little upgrade to some of your equipment would give you a huge difference in ppd/watt, so that might be worth looking into in your case. If you're really blowing circuit breakers, that seems crazy for the number of ppd you're making. For instance, I have an Atom 330 box, it makes about 950ppd, and it burns about 60W. I have another Atom 330 system with a different motherboard, it has an ION on the board and it has a pci-e slot I put an nvidia gts 250 into, and it's currently doing 6400ppd, and burning about 160W. Although even with a little cheapo core 2 duo, you'd get to a much better ppd/watt point, and it would definitely get you back into the 1500ppd range you were previously at. And come to think of it, at some point I'll probably upgrade my little core 2 duo system, do you want a deal on my old motherboard? I can't guarantee I'll upgrade it anytime soon, and probably not this month, but at some point it'll definitely happen. Let me know if you're interested.