P-90 years

[eatmadustch]

Hi
I can't understand why you chose P90-years in the stats ... I mean who's got a P-90 today? Why don't you use a more average CPU, like a PIII-1GHz or even a PIV-2GHz? I, and I'm sure many others would prefer that, as it's easier to compare with their computer and normal, modern computers.
Apart from that I really think the stats are perfect

[rshepard]

I would guess because it's sorta "traditional"-- I think GIMPS used that measurement way back when, and probably still does

[MAD-ness]

GIMPS still uses it.

You have to have some basis, and this is a good enough one.

Plus, when talking about CPU years, if you go with a 1 ghz P3, people will only do a couple cpu years per calendar year.

That would be a bit boring for stats.

[shifted]

I have my router running the sb client, and it's a P166mmx. So yeah, people do still use them.

Granted, it's just a tiny fraction slower than the Athlons i have running the client, but anything pentium or higher will finish a work unit in a reasonable amount of time.

It started k=27653, n=2918733 on Sun Nov 17 04:24:15 2002, and it's 91.82% done now at Mon Dec 9 15:33:12 2002. It'll go faster when it gets one of the lower n values for a different k.

[MDFaunce]

Originally posted by shifted
It started k=27653, n=2918733 on Sun Nov 17 04:24:15 2002, and it's 91.82% done now at Mon Dec 9 15:33:12 2002. It'll go faster when it gets one of the lower n values for a different k. [/B]

I thought the expiration time was 2 weeks?

Doesn't that mean that you are just wasting your time? Since the test expired 11/31 and has probably been re-assigned?

I stopped running my P/166 since it couldn't make the 2 week deadline running 24x7.

[kmd]

Expiration time is 2 weeks since last communication. It gets renewed each time you send a block.

[MDFaunce]

Originally posted by kmd
Expiration time is 2 weeks since last communication. It gets renewed each time you send a block.

Well.

That changes everything

[dudlio]

MAD-ness, think about what you're saying. "If we measured people's lifespans in years, then they would only live as long as the average lifespan, and that would be boring."

Boring? What's the point of a rigged stat though?

[rshepard]

What's the point of a rigged stat though?

I don't really think its a rigged stat. You can measure processing time directly, but its not very signifcant.
Consider: You are running a 1GHz box, and I am running two 200 Mhz boxes. For the same time period, it would appear that I am contibuting twice as much processor time, but in terms of actual work done, you are probably cranking out 2.5 times as much. The measurement of time gets indexed to some speed, so that you get an indication of processor power committed.

my $0.02 worth

[MAD-ness]

It isn't rigged.

It isn't bogus either.

It inflates the numbers, which isn't such a bad problem.

An example:

In GIMPS, I have been running a P4 2.0 Ghz CPU doing LL testing. It has completed two LL tests (primality testing, takes a LONG time). Each one took about two weeks (computer on 24/7) and it resulted in 1.6 P-90 CPU years EACH. So, in one month on a P4 @ 2.0 Ghz I am doing the equivalent of ~3.2 P-90 CPU years.

That is just a fraction of a P-90 CPU year per day. Not a very exciting stat. "Whoohoo! My stats went up .000034 today!"

Now imagine if the reference unit was, say, a P3 1ghz CPU year. You would get .000000lotsofzeros000034 CPU years (a totally arbitrary and inaccurate number I just made up) per day.

That is a lot of decimal places to keep track of and it makes the stats harder to read, in many ways.

One option would be to measure vs a modern CPU but make the time interval days rather than years.

Either way, it is a relatively arbitrary reference CPU, but it does the job and it is a bit of tradition.

I still maintain that it has integrity as a measurement of the actual computational power contributed (atleast in GIMPS), far more so, if properly applied, than cEMs.

[dudlio]

So your P4 is 30x faster than a P90. I'm sure you knew that

Anyway, it's cool. There were a lot of things about the stats that were confusing, and they've gone and fixed them all. P3-866's became Gigahertz, they stopped counting blocks and ranking teams both by tests and iterations. Now if they would only stop pluralizing cEM...

I'll concede the reference processor. I'm new to DC so I'm looking at it from an absolute perspective.