I've found that CPDN grants ~50% more credit per second of CPU time than Predictor does, on the same computer. Is there any way this bias can be fixed?

Across projects, my impression is that CPDN grants more credit than LHC, SETI, or Einstein as well.

See <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=1742">this thread</a>.

I commented in reply to your earlier post on the Questions &amp; Problems forum on why experiences vary on different machines. There is no way you can 'correct' the credits you are awarded - that would open up the way to endless cheating.

---

&gt;
&gt; I've found that CPDN grants ~50% more credit per second of CPU time than
&gt; Predictor does, on the same computer. Is there any way this bias can be
&gt; fixed?
&gt;
&gt; Across projects, my impression is that CPDN grants more credit than LHC, SETI,
&gt; or Einstein as well.

Have you checked out how much credit your hosts claim at the other projects? It may be higher than what is granted. One of the main differences from cpdn to the other projects is that cpdn is currently the only one that doesn't send the same wu to more participants and compare their results. This means if your host claims an amount of credit for cpdn work, that amount is what is granted - which is not the case per definition at the other projects. As far as I can tell from Andrews link, it would seems that on cpdn there is not even such a thing as claimed credit.

Well, in either case, cpdn is not very comparable with the others as it is very different in how it works.

---

&gt; See <a> href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=1742"&gt;this
&gt; thread</a>.
&gt;
&gt; I commented in reply to your earlier post on the Questions &amp; Problems
&gt; forum on why experiences vary on different machines. There is no way you can
&gt; 'correct' the credits you are awarded - that would open up the way to endless
&gt; cheating.
&gt;

Thanks--and sorry for posting in both places. As you say, it may be a platform-dependent thing, with some computers better suited to some projects. It may be worthwhile for me to compare more of my computers directly across more projects.

Or that may be a complete waste of time, since it's the science (no credits) that's important... ;)

&gt; One of the main differences from cpdn to the other projects is that cpdn is currently the only one that doesn't send the same wu to more participants and compare their results. This means if your host claims an amount of credit for cpdn work, that amount is what is granted - which is not the case per definition at the other projects.

That's far from accurate. CPDN does resent some WUs of re-check consistency among different platforms.
What it doesn't, is the concept of claimed/granted credit. Credit is granted upon actual work being done, not some synthetic benchmarks and others not-much-sayings numbers.
You finish whole model, you get 6805.26 credit. As simply as that; no horse-trading.

&gt; As far as I can tell from Andrews link, it would seems that on cpdn there is not even such a thing as claimed credit.
Yes. Claimed credit would make no sense on CPDN project.

&gt; Well, in either case, cpdn is not very comparable with the others as it is very different in how it works.
Nope, is more easy and clear than other projects as of god-damn credit. Every trickle you sent (10802 timesteps, 72x per model), you get about 94.5 credit. As simple as that, no trickies - just trickles.

The different amount of credit per day is, IMO, basically generated by different CPDN performance on different CPUs, e.g. P4 vs. AMD (to thank or to blame Intel Fortran compliler). It is a known fact that Intel P4s with HT can do a much more that AMD(64).

Generally, P4s are good for CPDN a SETI, AMD is doing quite good on Einstein.

---

<i>phpBB forum for CPDN, all are </i><a href="http://www.climateprediction.net/board">invited</a>

&gt; &gt; One of the main differences from cpdn to the other projects is that cpdn
&gt; is currently the only one that doesn't send the same wu to more participants
&gt; and compare their results. This means if your host claims an amount of credit
&gt; for cpdn work, that amount is what is granted - which is not the case per
&gt; definition at the other projects.
&gt;
&gt; That's far from accurate. CPDN does resent some WUs of re-check consistency
&gt; among different platforms.
&gt; What it doesn't, is the concept of claimed/granted credit. Credit is granted
&gt; upon actual work being done, not some synthetic benchmarks and others
&gt; not-much-sayings numbers.
&gt; You finish whole model, you get 6805.26 credit. As simply as that; no
&gt; horse-trading.

I wasn't too clear here ;)

What cpdn doesn't do, like the others, is to send out the same wu to several participants, wait for the participants to return the results, compare them, and then grant credit by evaluating claimed credit - that is granting the lowest value, the middel value or whatever scheme is chosen.

Any comparison that may be done between wu's does not in any way affect the credit system - and I view it from the credit systems point of view in this thread ;)

---