Re: How much time would it take a human to calculate a WU?
Posted: Thu Jun 19, 2008 5:38 am
Perhaps we can't really understand how fast the human brain could fold a Work Unit until we can simulate a human brain with computers. We're some time away from that goal, and even further away from simulating the brain with a machine that we could go buy for cheap and carry in our pocket. A spare, more or less.
How fast could all of FAH fold one WU? That is, if the whole shebang could run a single 110 pt WU instance in a huge SMP effort-
If a P4-C folds @ rate 1.27X clock speed in GFLOPS, then the 2.8 GHz benchmark machine is folding @ 3.556GFLOPS and then
the answer turns out to be slightly over 1/10,000th sec. I know, I know, a ridiculous concept, but our current 1.9+ PFLOPS is within the range of guesstimates as to the power of the brain, if we could get it all focused on math instead of girls, or whatever.
F@h achieved 1 PFLOPS sustained in March, 2007 with the implementation of the PS3 client. We've been running 1.9+ PFLOPS sustained for some time now and should reach 2 PFLOPS sustained in a few days to weeks, thanks to the GPU2 client.
This rate of increase is slightly faster than the (modified) Moore's Law of exponential technological progression, i.e. doubling every 18 months as we've doubled in about 15 months.
At this rate, we'll increase F@h performance by 3 orders of magnitude, reaching 2 ExaFLOPS in 12.5 years, but probably sooner than that, since the rate of increase also seems to be accelerating. Our speed increase above the standard curve is probably due not only to the fact that the device speed is increasing, but the number of donors and clients/donor is also increasing.
Will computers simulate the human brain within that 12.5 year time span? We'll see.
As an aside, we've added 80 TFLOPS since the nVidia GPU2 client became available, what, 36 hrs ago?
Ps Somebody out there is already successfully simulating the cerebellum...
How fast could all of FAH fold one WU? That is, if the whole shebang could run a single 110 pt WU instance in a huge SMP effort-
If a P4-C folds @ rate 1.27X clock speed in GFLOPS, then the 2.8 GHz benchmark machine is folding @ 3.556GFLOPS and then
the answer turns out to be slightly over 1/10,000th sec. I know, I know, a ridiculous concept, but our current 1.9+ PFLOPS is within the range of guesstimates as to the power of the brain, if we could get it all focused on math instead of girls, or whatever.
F@h achieved 1 PFLOPS sustained in March, 2007 with the implementation of the PS3 client. We've been running 1.9+ PFLOPS sustained for some time now and should reach 2 PFLOPS sustained in a few days to weeks, thanks to the GPU2 client.
This rate of increase is slightly faster than the (modified) Moore's Law of exponential technological progression, i.e. doubling every 18 months as we've doubled in about 15 months.
At this rate, we'll increase F@h performance by 3 orders of magnitude, reaching 2 ExaFLOPS in 12.5 years, but probably sooner than that, since the rate of increase also seems to be accelerating. Our speed increase above the standard curve is probably due not only to the fact that the device speed is increasing, but the number of donors and clients/donor is also increasing.
Will computers simulate the human brain within that 12.5 year time span? We'll see.
As an aside, we've added 80 TFLOPS since the nVidia GPU2 client became available, what, 36 hrs ago?
Ps Somebody out there is already successfully simulating the cerebellum...