While looking for some P4s to see how they did - I noticed Avi's 2.9Ghz Celeron. Claimed credit= 20.7 points. Granted credit= 47.9. (Others were closer to 1.5 times the claimed credit.)

If this holds true for the rest of the P4 line, then those who were discouraged by the disparity of the P4 line versus the AMD cpus may have had their prayers answered now.

But the range of scores on the Celeron bring up the question that's been asked before: Is the credit/model score being based on the same consistent set of systems, (thus various algorithms used in the WUs operate faster than others on the Celeron) - or is it based on an average of whatever machines were available at the time on Ralph? (With higher scores when larger numbers of optimized clients were represented.. and lower scores when larger numbers of P4s on Linux running the standard client were represented?)

---

I just noticed a seriously wrong credit/work value being used. For some reason, my script gave extremely low values for a handful of work units so as a quick fix, I reset the credit/work values to adjust as results come in. Sorry about this mishap. I'll have to do some damage control this morning but now I must sleep.

We may actually have the credit/model values adjust for all work units as results come in and use ralph tests to serve as a starting point.

I think this is an excellent idea - perhaps the sample sizes on Ralph were really a little bit too small to determine reliable averages...

---

Team betterhumans.com - discuss and celebrate the future - hoelder1in.org

We may actually have the credit/model values adjust for all work units as results come in and use ralph tests to serve as a starting point.

I think this is an excellent idea - perhaps the sample sizes on Ralph were really a little bit too small to determine reliable averages...

The low score I noticed had a very high number of decoys generated, it's possible that because each decoy was so quick, there were rounding issues in the inital data collection causing the credit to be too low?

But overall, once the glitches are worked out, it's looking good. I think even the people unhappy with the new system at the moment will come to like it given time, since it gives a very clear way to compare the performance of an Ubermachine :-) (Whether highly overclocked or massively parallel).

---

Why do a lot of Work units with long computation time have fewer Credits than others with short computation time?
Longer computation times mean nevertheless more Credits. Or?
Otherwise the Creditsystem would be garbage.

Athlon

---

I just noticed a seriously wrong credit/work value being used. For some reason, my script gave extremely low values for a handful of work units so as a quick fix, I reset the credit/work values to adjust as results come in. Sorry about this mishap. I'll have to do some damage control this morning but now I must sleep.

Would that explain this:

""Computer ID 275524
Report deadline 22 Aug 2006 19:21:18 UTC
CPU time 85019
stderr out <core_client_version>5.4.9</core_client_version>
<stderr_txt>
# random seed: 2986845
# cpu_run_time_pref: 86400
# DONE :: 1 starting structures built 38 (nstruct) times
# This process generated 38 decoys from 38 attempts


BOINC :: Watchdog shutting down...
BOINC :: BOINC support services shutting down...

</stderr_txt>
Validate state Valid
Granted credit 160.321852290615
Granted work credit 0.9088843809044
application version 5.25 ""

Just wondering ;-)
Francis

---

That doesn't look right at all! I'm sure you'll find that the work credit suddenly jumps upwards once he's had the chance to rerun his scripts.

---

David Kim explained this morning that he made a mistake in his scripts so the Granted work credit is wrong for a few of WUs. This could explain that.

Yes, these extreme differences are due to the script error. I will fix them soon and adjust the work credits. Sorry about that. The smaller differences when considering the credit/cpu time are due to the random nature of rosetta. The same work unit with different random number seeds will have different search paths so they are likely to have different credit/cpu time values when comparing small numbers of runs. The values should average out and in my opinion the differences are acceptable.

You had a good (short) night? ;-)

Yes, these extreme differences are due to the script error. I will fix them soon and adjust the work credits. Sorry about that. The smaller differences when considering the credit/cpu time are due to the random nature of rosetta. The same work unit with different random number seeds will have different search paths so they are likely to have different credit/cpu time values when comparing small numbers of runs. The values should average out and in my opinion the differences are acceptable.

So setting the work unit length to 24hrs would give more or less credits than the standard 3hrs?

---

So setting the work unit length to 24hrs would give more or less credits than the standard 3hrs?

It should give 8 times more credit per WU, and the same credit per hour in a perfect world.

---

Grüße vom Sänger

So setting the work unit length to 24hrs would give more or less credits than the standard 3hrs?

If you are talking about the particular work unit, more credits because you will generate more models, but if you are talking about overall what the best work unit length is for maximum credits, there are a number of factors that can come into play and I don't know the answer.

Granted credit 27.5148233945073
Granted work credit 35.3874794761889

Just wondering which of these two scores I'm receiving.

---

98SE XP2500+ @ 2.1GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

both, but the granted credit is what gets reported to the stats sites.

Granted credit 27.5148233945073
Granted work credit 35.3874794761889

Just wondering which of these two scores I'm receiving.

This is the way i hear its "supposed" to be currently.

There are currently two ranking systems. One is the traditional, Granted Credit, which is exported to the ranking sites. The latter is the new one being tested. From what I hear, those using stock apps and are linux or P4 might be getting higher. Those using optimized apps should be getting lower. What the future plans are about KEEPING both, Im unsure, but it provides a way to see that the new credit system is "working right." The new RAC on this site should soon be sorted by the new, "Granted work credit" if its not done already.

---

Thanx for the info , was just curious.
I think I like the new credit system better, hope it gets adopted soon.
:)

---

98SE XP2500+ @ 2.1GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

I don't know what to think of this new crediting system. I'm running offline with a vanilla Linux client, which in the past kept receiving lower benchmark scores than did Windows clients. So far, I've reported to the new system (all at the same time, when I dialed in this Thursday) the completion of four WUs. The respective numbers, granted credit (old) on left, work credit (new) on right:
66.4 : 055.8
66.6 : 056.0
68.8 : 149.4
68.6 : 159.2

Is it just a coincidence that the two high (new) numbers are for WUs I reported on the day their deadline would have expired, whereas the two low (new) numbers are for WUs whose completion I reported ahead of their deadline date ?

And I am disappointed that there are two WUs on which the new crediting system will give me 16% lower numbers than the old system (which was already giving me lower numbers than for many other participants).
.

I don't know what to think of this new crediting system. I'm running offline with a vanilla Linux client, which in the past kept receiving lower benchmark scores than did Windows clients. So far, I've reported to the new system (all at the same time, when I dialed in this Thursday) the completion of four WUs. The respective numbers, granted credit (old) on left, work credit (new) on right:
66.4 : 055.8
66.6 : 056.0
68.8 : 149.4
68.6 : 159.2

Is it just a coincidence that the two high (new) numbers are for WUs I reported on the day their deadline would have expired, whereas the two low (new) numbers are for WUs whose completion I reported ahead of their deadline date ?

And I am disappointed that there are two WUs on which the new crediting system will give me 16% lower numbers than the old system (which was already giving me lower numbers than for many other participants).
.

From what I have read in other threads, there was a problem with the new credit system last night but it was caught quickly and corrected. Also read that they would be correcting the ones that are wrong. You just happened to submit a couple results during that time. The new system when fully implemented will be very good for you.
:)

P.S. I found this very interesting link to an informative page.

---

98SE XP2500+ @ 2.1GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

P.S. I found this very interesting link to an informative page.

Whose site is that? It looks and sounds official, yet its not the correct url.

---