Many have tried...

all have failed.

Try working it out yourself. The result you get is likely to be no better or worse than previous attempts.

I'd start off with the CPU hours turned in the last few weeks, take a days performance and indeed do a stab at the FLOPS a single CPU turns in on average, then compute the seconds of a days production and multiply that by the CPU performace......BOINC time is real time, UD agent time is clock time, so you'd have to guess how much slush sits in that mix too.....enjoy

PS I'd be disappointed if there is not a Petaflop lurking there.....

If it is then would it not make more sense to just measure it with a pedometer? Another important consideration... doesn't the act of calculating/measuring such a thing marginalise all the people who wear no shoes at all? And what about the animals? I imagine the centipedes would have a lot to say about the matter.

P.S. I just received a nasty email from K.D. Lang. She doesn't appreciate talk about PETA flopping.

Well, I can't say that's a bad estimate but I can't say that's a good estimate either. Nor can I say I have a better estimate or any idea how to make a better estimate. All I can do is point to the inaccuracies in your estimate.

First, the science apps that do the crunching are not optimised for SSE3 so obtaining your Sandra benchmark with SSE3 introduces inaccuracy right off the top.

Second, the WCG work units (WUs or tasks) vary greatly in length and complexity therefore the 132 TFLOPS per task may be far from average.

Is your estimate of ~200-230 in the ballpark? I like to think so but I bet others who have greater expertise can offer convincing arguments that it's not even on the same street as the ballpark.

Just for Fun: 62,480.700 GFLOPS .........Voux: Pays Bas .... Douze points

PS....this was a 1 degrees Kelvin, superconduction cooled calculation ;>)