LHC@home

> Oh, I screwed up then (math in public *g*)

Why do you think I left it up for someone else?

> So make that 5.57 TFlops, that's lower than I expected but likely at least
> closer to the Ballpark.
>
> I would estimate the actual number around 10-15 TFlops as of now.

Which is still not that bad. We are way low on participants still. There are oodles still running SETI@Home Classic. When that is shut down, I don't see how UCB is going to keep up with the demand for work. I know they will be able to re-allocate their computers and should be able to free up some additional hardware for the servers.

But I still don't see how they are going to feed the masses.

---

> Oh, I screwed up then (math in public *g*)
>
> So make that 5.57 TFlops, that's lower than I expected but likely at least
> closer to the Ballpark.
>
> I would estimate the actual number around 10-15 TFlops as of now.

Falcon,

If you believe benchmark scores (and who does ;), then your original number is...more correct than the 5.57.

But what does it all mean anyway....

Seti WUs claim they are 27.9x10^12 FP ops of work, but really are closer to 5x10^12. (Their test_with_me WU is 5.78...as tested by Intel's Vtune program...but it usually takes about 25% longer than an average WU)

Einstein still claims 20x10^12, but thats a holdover from back when they had short WUs.

Since BOINC has corrected its benchmarks (forcing the compiler to not remove chunks of code, as the results of this computation were never used) I doubt the scores match results of other benchmarks.

They certainly don't predict completion times for any of the projects WUs very well...tricky to do really.

So...which parts do we decide to "take on faith?"

Cobblestone - based on benchmarks?
Credits - based on averaged WU claimed credits (or lowest because of 0 claims)
Original classic seti's FLOP claims (I know they were being conservative)

Bottom line...
100,000+ CPUS of any kind can crunch pretty quickly.

---

> Bottom line...
> 100,000+ CPUS of any kind can crunch pretty quickly.

Good points made there, I'd sign that statement :)

---

Scientific Network : 45000 MHz - 77824 MB - 1970 GB

> Bottom line...
> 100,000+ CPUS of any kind can crunch pretty quickly.

And 6,000 and change can too ... :)

LHC@Home has kept a limited number of Participants and we still run them dry pretty quickly.

---