Hello Nite Owl: This doesn't appear to be a WCG issue but a BOINC issue and I am kicking myself for having a short memory.
We saw this same issue last April-May in Rosetta that the Intel based machinery got shafted in terms of points compared to the AMD's. That was my orginal reason for using crunch3r's files back then. They fixed that issue albiet at a higher level.
When we came to WCG with all the new issues concerning the quorum this totally slipped my mind. Have spent the last hour analyzing my own point claims versus others and now know why with equal runtime I am 40+% behind lesser AMD machines.
This is no more acceptable to me than how the Linux people feel.

Zombie:
Good point. The problem is that one size does not fit all. I'm completely in favour of a change but it will need lots of research. The problem comes down to the different architecture of the CPUs, the different demands placed on them and their response to those demands
You will find, if you look carefully, that there is a marked performance difference, on the same machine, between it's performance on FAAH and HDC. This will always be the case. WCG runs a variety of projects, all placing different performance demands on the CPU and each CPU architecture responds differently to those demands. A system put in place for one project type is unlikely to be a good solution for another. Likewise for CPU type.

I have tried a few times to gather relevant data in the hope of getting to the root of this problem. There has been little response. We also need to keep in mind that as and if projects with science optimizations come on line it will cause a whole new set of similar problems only worse.

The BOINC point system may have been sort of OK when it was developed but has effectively become little more than a blind auction a great theory that simply doesn't work in the real world.. That situation has been improved, by the latest initiative here but doesn't begin to address the root cause. Having said that, what will? I've had my say in other threads, how about getting some of the keen crunchers and IT professionals in your teams to brainstorm a workable solution, on a multi platform, multi architecture, multi project basis and submit it for review.

WCG has some very unique problems awarding points on a fair basis given its unique structure. Complaints wont help. I'm sure they know by now. WCG have already shown us that they will look at hard evidence, properly presented to them and a willingness to act. Those who are capable should be offering workable, documented solutions. I would suggest starting with a clean sheet of paper and go from there. Keep the recycling bin handy you'll need it Don't expect rapid response. The solutions would only be a rough candidate for Alpha status at best. But you will have done something positive and constructive and may even sleep just a little better.

Cheers. ozylynx

This is a very rough and unchecked draught of a idea I had since my previous post this morning.
It may be totally unworkable and certainly not completely thought through. Take a look anyway.
I'm not a mathematician so forgive my stumbles please.

P(x) = Project C(x) = CPU Family B = Baseline CPU

If computer B produces Y(B) points/hour on P(a) and runs for X hours to complete a WU for P(a)
Our base score becomes:
X*Y(B)= total points P(a) or TP(B). World Community Grid would need to use the same computer to represent B as was used to calculate existing stats such as CPU days etc.,

Y(B) will be a constant. X is a variable

Computer A, on the same work unit, produces Y(A) points/hour on P(a) and runs for X hours to complete a WU for P(a). The result would be :
X*Y(A) = TP(A)
We can now calculate:
TP(A)/TP(B)*100 = A, performance factor for P(a) or A(pf)%. A constant. We now have a direct comparison of performance between our benchmark machine and the reporting machine for this project.

Any further work performed by computer A on P(a) would be calculated by the formula:
edit: X(A)*Y(B)*A(pf)% = Points claimed =TP(B) (theoretical) {End edit}

This formula should then be valid for all other C(x) of the same type on the same project. It takes into account the different performance of the CPU architecture as applied to a specific project and leaves time as the only variable. It can be seen that the score for machine B should exactly equal to the score for machine A and that if machine A is faster than B it will produce more results over time and be effectively awarded more points on a direct linear scale. Variable size of individual WU would have the effect of raising or lowering X, our variable, and points claimed would reflect this variation.

A database would need to be established for each basic computer architecture C(x).
Computers, upon joining the project would need to accurately report their CPU type for comparison with the Dbase and previously unknown types would need to have comparison testing done for each Project type. i.e. (pf)% would need to be established for individual science projects.

The process would, need to be done for each science project and a P(x)(pf%) number be produced for each one. Any machine which is benchmarked this way, against B could then be used as a benchmark machine in a quorum or Beta, as that machine will have a known (pf)% so there would be no need to directly benchmark against B after the first few are completed.

Outliers should be eliminated and any that do happen will be instantly identifiable.

This is an unwieldy method requiring plenty of testing, cross referencing and of course a flexible database. Not to mention the CPU reporting problems, which already exist would need fixing.
It just might be a starting point for another idea however.

Cheers. ozylynx

Carl wrote:
About time too....

Ozylynx, where would you be getting the "points/hour" figure from ?
Would that be based on the same flawed BOINC benchmark we are using now, which penalises certain CPU types ? If so, I dont see this being any better mate. What your formula seems to do to me mate, is start off flawed and then take points off some people and redistribute them to others.

Dont really know what the answer is at this point myself though. But as long as anything is based on the current BOINC benchmark, I cant see it working properly.