An order of magnitude is a pretty significant. Estimates are fun and interesting but obviously we can't rely on them very much.

Ranking supercomputers is interesting too. Here they are ranked by performance on the Linpack benchmark. By that measure IBM's BlueGene/L seems to be the reigning champ. Here one of the BlueGene's competitors in 2004 questions the Blue Gene's efficiency. In the end what matters is how fast it crunches the data. If lots of CPU cycles are spent on system overhead then the machine isn't really as fast as the Linpack benchmark indicates it is. But the BlueGene's benchmark is so much higher than its competitors' benchmarks it seems that even if it were very inefficient it might out crunch it's competitors anyway.

I wonder what a BlueGene/L costs?

An average of FLOPS needed to compute a task. Yep. That would allow a decent estimate. So where are we going to get that number?

Yes, Jackson. So many toys to choose from; so little cash

Thanks, Didactylos. 3e13 FLOP is the number we've been lacking, the number that makes a reasonable estimate possible. Repeating the first calculation using my computer's benchmark and CPU times for a few WUs indicates 3e13 is pretty close.

Since the 70300 results per day is an average going back to the beginning of WCG it doesn't reflect today's power accurately. The stats show results per day has been over 100,000 almost every day since end of November, 2005. There are lots of 150,000 days. It looks to me like 130,000 results per day is more like the average over the past 6 months rather than a peak.

Well, I'm tossing out bugsan's 230 estimate because it rests on very shaky assumptions.

I don't know how Sekerob's "real daily time angle based on last month's performance" works. Show me the formula/calculations and I might buy it.

Didactylos provided enough info in his post to see exactly how he calculated his 24 and 45. I assume he got the 3e10 number from reliable sources inside WCG so I accept 3e10 FLOP as the size of the average WU. I'm tossing out his 24 because it's based on an average that goes way back to the start of WCG and does not accurately reflect current computing power.

Today I totalled the results for the past 30 days (June 22 to May24), divided by 30 and get 130,167 for the average daily results, round off to 130,000. That simply confirms what I estimated in my previous post so I still maintain 130,000 results per day is not a peak, it's current average daily production which means 45 TFLOPS is a very reasonable estimate of current computing power.

Sekerob, show us how you calculated your 62, please.