Over the summer, for kicks, I figured out what it would take to build a machine capable of crunching a million credits per day. This machine turned out to be a rack filled with high density server chassis (2 server MBs per 1U chassis, each with 2 hex-core, hyperthreaded Xeon CPUs.) That's 48 threads per every 1.75 inches of rack space. There would be about 20 or so of those, and along with the requisite rack mounted UPS and routers, it would cost something over $100K. That's not including running the extra 100 AMP power circuit to my house. (In retrospect, it might work out better using Opteron CPUs.)

This was before the Ken's CUDA ppsieve application was released.

Right now, you can build a mega-credit computer for about the cost of a Windows gaming laptop or a mid-range MacBook.

A modest SLI-capable computer with a pair of GTX 580s should be able to crank out about a million credits per day, for a total costs of less than $2000.

Mind-boggling, isn't it?
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

It's completely understandable, but the downside to having the GPU version be so much cheaper than the CPU version is that it will only be able to crunch a million credits per day on a limited number of projects, as opposed to any project that you want to crunch a million credits per day on.
____________

You'll also never find a prime with a GPU megacruncher:)

The work throughput GPUs offer really is amazing compared to CPUs, where supported e.g. in sieve. But I'm still waiting for the one that'll do the same for LLR.

I guess it depends on what you're after, but personally boinc credit isn't terribly important to me except as a side effect to collecting the badges here. In pure credit terms, I have a single GTS450 on collatz scoring almost 50% more boinc credits than the rest of my farm here: equivalent to 40 GHz single core modern CPU + two older GPUs thrown in.

If I do have a scoring goal, I'm trying to improve my ranking on the world top prime finders list... so just as well the GPU can do some sieve else I wouldn't do much at all.

Does this greater speed mean that the optimum sieve depth has changed?

Just to be clear, what I did was spec out such a machine. I didn't actually build it. Believe me, if I had a spare $100,000 sitting around I have much better things to spend it on. (I did, however, conclude that it could essentially eliminate my heating costs in the winter, but would substantially increase the air conditioning bill during the summer.)
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

DP
____________

Nope, but you will find a whole lot of composite numbers! :)

True...but I would gladly trade in my ~1.5M composites found so far in the PPS subproject just to be a wingman on the next SOB prime found!

It is pretty awesome though that in the next week I will have accumulated the same credit that it took me two years to accrue previously via CPU...and that's only with a pair of 9800 GTs. I can't wait until I can figure out how to get my two ATI cards working here!

NOTE: I have the Million Credit Per Day Machine here already, my Dual GTX 580 Box should be able to hit about 1.25 Million+ as long as the tpsieve Wu's keep flowing. But I really built the Box to run at the GPUGrid Project when I discovered I could run it here too so I turned it loose for awhile on the ppsieve/tpsieve Wu's for awhile. Probably until after the next challenge anyway ...

No I didn't read that, oh well I'll get what I can get and more over to the GPUGrid Project with um if I have to ...
____________

I am sure there will be challenge work available. Even the current n<3M will not be optimal by then IMHO, and even if I'm wrong, there will be 3M<n<6M to work on by then.
____________

NOTE: I have the Million Credit Per Day Machine here already, my Dual GTX 580 Box should be able to hit about 1.25 Million+ as long as the tpsieve Wu's keep flowing. But I really built the Box to run at the GPUGrid Project when I discovered I could run it here too so I turned it loose for awhile on the ppsieve/tpsieve Wu's for awhile. Probably until after the next challenge anyway ...

Did you read the thread PPS (Sieve) changing gears again?
I believe there will be no work for this in the next challenge because PG reaches the optimum sieve depth much earlier...