The Seti project has at last been beaten as the most powerfull computer system on the planet

http://news.bbc.co.uk/1/hi/technology/3983131.stm

The new supercomputer chugs along at 70tf, the original seti project is currently a meagre 68tf

---

<img src="http://www.boincstats.com/stats/banner.php?cpid=f098ec1961b5cdf855cbac04a444071f">

ok not the fastest BUT normal people can be a part of it !
at the big servers only the super admin living in a ice palace dream her dreams of magic.;-)))

this nice thing about SETI is that it is very easily expanded. As more and more people with more and more machines come into the fold, our computer power will increase. Whereas, the update the supercomputer would cost a whole lot of money and be a major undertaking.

Jim

> The Seti project has at last been beaten as the most powerfull computer system
> on the planet
>
> http://news.bbc.co.uk/1/hi/technology/3983131.stm
>
> The new supercomputer chugs along at 70tf, the original seti project is
> currently a meagre 68tf
>

---

> The Seti project has at last been beaten as the most powerfull computer system
> on the planet
>
> http://news.bbc.co.uk/1/hi/technology/3983131.stm
>
> The new supercomputer chugs along at 70tf, the original seti project is
> currently a meagre 68tf
>

I think it's more fair to compare that system against BIONC as a whole, once S@H 1 is turned off.

---

Luckily our ~65 TFlops are 'sustained' and on a real-life application.

Those Supercomputers usually deliver only 1/2 to 1/4 of their rated 'peak' performance in real-life applications.

Anyway, guess we just need a bit more power to get the relation of numbers back where they belong with SETI and/or BOINC taking No.1 spot again ;)

---

Frankly I think everyones calculations are off/mistaken.
But I can play that game too ;)

Currently in BOINC, according to boincstats.com, had a recent daily combined total credit of 7,494,867 credits.

From the definition pages of BOINC and Paul's "unnoficial boinc documentation", a credit is...

"A Cobblestone is 1/100 day of CPU time on a reference computer that does:
1,000 double-precision MIPS based on the Whetstone benchmark.
1,000 VAX MIPS based on the Dhrystone benchmark."

So, how many Teraflops = how many credits...you ask ;)

Cobblestone reference machine does One Billion FP Ops per second (1000 MIPS).
In a day a cobblestone machine will do 100 credits (running full time, etc.)

There are 60 * 60 * 24 seconds in a day or 86400. Divide by 100 and you have 864 seconds.

Therefore a credit would seem to be, the ammount of computing power a cobblestone machine can perform in 864 seconds.

864 * billion = 864 Billion FP Ops = 1 credit....
It would take 1,000 cobblestone machines to generate 1 Terraflop per second.

Now lets turn credits / day into FP Ops per day.
7,494,867 credits * 864 billion FP Ops per credit = 6,475,565,088,000,000,000 FP Ops per day

[bignum] / 86400 = 74,948,670,000,000 FP Ops per second

divide by a trillion and... 74.95 Teraflops per sec

Seti@home by itself had 5,436,792 daily credits...and therefore is running at 54.37 TFLOPS.

S@H used 125,183 CPUs to do this (accoding to boincstats), so really the average machine seems to be about half of a cobblestone machine.

Is this a reasonable number?

Claimed:
Folding at home has 169,180 active CPUS and claims it has 195 TFLOPS
-- http://vspx27.stanford.edu/cgi-bin/main.py?qtype=osstats
(I suspect the folding people actually know how many FLOPS are in their WUs, and they also have SSE code to improve speed)

While Project Columbia is armed with 10,248 Itanium 2 processors and achieves 50 sustained teraflops.

IBM's Blue Gene/L prototype (they are still building it) "...has over 32,768 customized processors, actually modified variants of the dual core POWER-based (Power PC) architecture." Currently they claim 70 teraflops.

"...Virginia Polytechnic Institute, which hit the No. 3 spot in 2003 using 1,100 Apple G5 computers, each with dual 2.3GHz processors." They are upgrading.
"...The original System X operated at 10.28 teraflops for the official records, but its peak theoretical performance was rated at 17.7 teraflops."

So with 125K+ CPUs yea, 54 TFLOPS doesnt sound too overly large.

But thats just BOINC. Doesn't include original Seti@Home.

Original Seti@Home claims 1,518,043 results in last 24 hours representing 5.920368e+18 FP ops in that time, or 68.52 terraflops/sec. Frankly I think they are underestimating. With 450,000+ users returning at least one result in last 28 days, that should represent at least 450K+ CPUs of some speed or other.

Also, lets not forget Seti uses Whetstone benchmark and other super computers use Linpak, so who knows.

---

^ Bump.

Any of my math wrong?

There are lies, damned lies, statistics, and then there are benchmarks and TF ratings. The only way to accurately measure the "speed" of any system is by pitting each system against each other in various specific fields. Ex.: Deep Blue plays an excellent game of chess, and Earth Simulator Center can't, and both suck at protein folding. Now, who's the better machine to sun SETI?

There was a reduction in the meaning of a CS by 300 to make the numbers for short results visible.

---

BOINC WIKI