What is the fastest time for an "average" workunit?

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

DaveT - 24 Jun 2009 07:56 GMT

Looking at 14 completed workunits, I can see that the elapsed times range between 1-4 hours, with
some as long as 7 1/2 hours. The actual times for the non-cuda units are 1:03, 1:01, 3:33, 3:23,
1:03, 1:07, 3:31, 3:44, 3:36, 3:50, 3:42. For the cuda units, they are 2:31, 2:31, and 7:28.

Ruling out use of the machine for other things, why do some units take so much longer than others?

Based on these and other observations, I would say the "average" time for an "average" unit on my
2.5 ghz 45 watt cpu is 2 1/2 - 3 hours. Because it is dual-core with cuda, it actually processes 3
units during that time.

I was wondering, what is the fastest time for an "average" unit on the fastest processor? Note, I
mean a single core processing single unit, although this might include some cryogenically cooled lab
machine with a scillion math coprocessors or some power guzzling gamer thing.

I'm not looking to start a war here, I was just curious as to what is out there. It is near my tenth
anniversary, and just for the heck I ran a (current) workunit on one of my original computers, a
55mhz cyrex. It took about a week.

Reply to this Message

Tazz - 24 Jun 2009 14:34 GMT

> Looking at 14 completed workunits, I can see that the elapsed times range between 1-4 hours, with
> some as long as 7 1/2 hours. The actual times for the non-cuda units are 1:03, 1:01, 3:33, 3:23,
[quoted text clipped - 13 lines]
> anniversary, and just for the heck I ran a (current) workunit on one of my original computers, a
> 55mhz cyrex. It took about a week.

IIRC,the crunching time is affected by the AR (angle range) of the unit.
That is, a unit with an AR of 1.2 will be faster than a unit with an
AR of .12 http://www.boinc-wiki.info/SETI@Home_Enhanced_FAQ It's a bit
outdated but it'll give you a rough idea.

I have about a hundred original SETI@Home (SETI Classic) units. The
command line app with SETISpy and SETI Driver. I should run it on a
modern computer to see what kind of times I get.

Signature

</Tazz>

Reply to this Message

DaveT - 24 Jun 2009 18:01 GMT

>> Looking at 14 completed workunits, I can see that the elapsed times range between 1-4 hours, with
>> some as long as 7 1/2 hours. The actual times for the non-cuda units are 1:03, 1:01, 3:33, 3:23,
[quoted text clipped - 22 lines]
>command line app with SETISpy and SETI Driver. I should run it on a
>modern computer to see what kind of times I get.

Thanks. I didn't know about this page. They don't go into much detail about angle range. Like some
of those irritating help files, they mostly say "the angle range is the range of the angle". Like
what, if its pointing straight up the angle is 90, if it's pointing sideways the angle is zero, or
whatever. And why, perhaps, this affects the crunching time?

Anyhow, what I was mostly curious about is what are the fastest computers processing workunits? Not
vast numbers of processors processing vast numbers of units, but a single processor processing a
single unit. Are there special computers optimized to run Seti, perhaps with more or faster floating
point processors, SRAM, or large caches? I have head of liquid nitrogen cooled CPU's running at
6gHz. Possibly some insanely large scalarity? Perhaps some engineering marvel at IBM, MIT, Intel or
in Japan or Russia that can chew through a workunit in thirty seconds? I just thought I'd ask.

Reply to this Message

Space Forum / SETI / June 2009

What is the fastest time for an "average" workunit?