[NCLUG] Power consumption and performance of Folding at Home.

[Sean Reifschneider]

I was looking at the client stats for F at H today and noticed that it was at
just a hair under 4 PFLOP.  Earlier this year it passed 1 PFLOP.  But I
didn't know how this compared to other supercomputers...

According to the top-500 list:

   http://www.top500.org/list/2008/06/100

According to that list, F at H is something close to 4x faster than the
fastest supercomputer (on that list, no telling what the secret computers
have) the Roadrunner at LANL.  It's 10 times faster than all but the top 3
computers on that list, and it's faster than the top 20 combined.

Then I noticed that the Roadrunner uses over 2MW of power, and I seemed to
recall that John was saying something about F at H using this much power...  I
went and looked at that Wikipedia page he referenced and noticed it says:

   [...] the current Folding at home project, if it were theoretically using
   the most efficient CPUs currently available, would use at least 2.8
   megawatts [...]

It also says:

   "the equivalent of 0.5-3 electrical wind mills depending on their size"

which I found fascinating.  But the next paragraph goes on to say that
estimating the actual power consumption of F at H is hard because many of the
machines would be running anyway.  That is true for me of 14 out of the 15
machines I'm running it on, though one of those could probably be suspended
when we weren't using it because I only rarely access it remotely.

Also, just for comparison, the Roadrunner cost over $133 million.  So, for
scale, F at H is like a half billion dollar supercomputer.  Which, according
to the F at H stats on number of currently active CPUs would mean that for
Stanford to reproduce the system at their own facility they would need
every current contributor to donate $1500 per CPU they currently run.  I'd
have to donate over $23k to make up for my contribution
(532,000,000/342,861*15) to F at H.

It's not clear from what I read whether the $133m for Roadrunner included
the cost to for 6,000 ft^2 of data-center space (300 racks), or if the
2.3MW it takes to run it includes the cost to run the A/C.  Ideally it
should.

To me, it seems like F at H is pretty economical.

Oh, and almost exactly as I write this, the Hacking Society team (team
number 58381, hint hint :-) has entered the to 0.3% of teams (counted by
number of points contributed since the start of F at H), which I think is
pretty cool...  We're at 430th place in all-time rank, and 163rd place
for points over the last week.

Sean
-- 
 If we don't survive, we don't do anything else.
                 -- John Sinclair
Sean Reifschneider, Member of Technical Staff <jafo at tummy.com>
tummy.com, ltd. - Linux Consulting since 1995: Ask me about High Availability