IIRC, I started this thread because I saw that for the same AR WU - ~ 10 %.

Finally.. a mistake in the 'credit adjust script'.

Then the credits increased again.


Maybe this time also?


If I look to other projects.. I don't say which.. I guess it's known which projects pay much much more..

This is like an inflation of the credit system.


AFAIK, ATI GPUs are now welcome at SETI@home because of the kind opt. crew.
So, maybe more advertisement for to recruit more members..?

OTOH, not enough AP WUs available.

---

IIRC, I started this thread because I saw that for the same AR WU - ~ 10 %.

Finally.. a mistake in the 'credit adjust script'.

Then the credits increased again.

Maybe this time also?

Have you read the script, or the explanation of how it works?

Because Multibeam in particular does not grant a linear amount of credit (because the kind of floating point operations change with angle-range) a run of "high paying" or "low paying" work will cause the overall credit to adjust.

That isn't entirely what we'd like, but the good news is that whenever it adjusts away from the correct value, it'll adjust back later -- and in the longer term, we'll be paid what should be paid.

---

Have you read the script, or the explanation of how it works?
[...]

AFAIK, IIRC, because the GPU cruncher were counted also. So we got a deflation.

Then change of the script (only CPU PCs) and the credits increased again.

---

Have you read the script, or the explanation of how it works?
[...]

AFAIK, IIRC, because the GPU cruncher were counted also. So we got a deflation.

Then change of the script (only CPU PCs) and the credits increased again.
[/color]

The script hasn't been changed since September 2008 (at least the one in the BOINC source), so considers only CPU time. The way it works ignores the extremes at either end anyhow.

With about 293 thousand active hosts with a total RAC near 72.9 million, the average RAC is slightly under 250. I believe the median host would have a lower RAC than that, the average is unduly influenced by the few hosts with very high RAC. IMO, the median host isn't running CUDA so variation of the credit rate is still strictly related to the mix of ARs being split.

David Anderson's latest approach to a revised credit system is CreditNew. Some test code was checked in nearly a month ago, commented with "It seems to work!". It will be interesting if it is put into actual use.

                                                                 Joe