Hi, I'm just starting in the world of linux and I've installed ubuntu version 9.04 but boinc package in it is only version 6.2.18 so I decided to upgrade to 6.4.5 or 6.6.20 but I came across the same problem in both cases.

That is the performance test gives me about 5000 integer MIPS per cpu under either 6.6.20 or 6.4.5 but about 8600 under 6.2.18 and this difference of performance does effect my calculations (I've calculated my ratio points/hour/core under all versions and it goes in the same way).

Could it be my computer dealing with energy options even if I've selected the "performance" mode? Has someone else came across the same problem?

Thanks for your help

PS: I'm French so if something isn't quite clear make me know...

It is just the way the Ubuntu distro of boinc is compiled. They compile with some optimizations enabled which causes the benchmarks to be higher than the Berkeley version which has no optimizations enabled.

I haven't used Linux recently, but when I used Ubuntu I noticed the same difference, and I don't believe it had any effect on the speed of the actual data crunching.

But the problem is that it actually has some effect on the speed of data crunching :D

I've verified it on already 2 projects that are Ibercivis and Geneticlife, for example for the former I have 26points/hour/core on 6.2.18 but 15points/hour/core on 6.4.5 and 6.6.20.

Maybe should I try other versions of boinc manager to see.

What you're talking about there, is about credits per unit of time. This is NOT the same as data crunching speed for the science application, which doesn't use the benchmarks.

Ok excuse me I didn't quite understand what you were explaining. In fact you are saying that the workunits take the same amount of time (that is the speed of data crunching is the same) even if they don't get the same credit per unit of time, that's it?

(So the credit per hour depends on the performance tests realised by the boinc-manager and not on speed of crunching??)

(So the credit per hour depends on the performance tests realised by the boinc-manager and not on speed of crunching??)

Basically, yes. But faster crunching will also affect it, as it's time taken times benchmark, or something similar.

If you compile BOINC yourself, it's apparently possible to change things a bit, thus giving a higher benchmark, which will give more credits for the same work done.

But some, (a lot?), of projects are moving to fixed credits, which don't rely on the benchmarks at all. So for these, it doesn't matter what the benchmarks show, as far as credit goes. But they are useful for comparing different computers, (provided they're using the same version of BOINC), to see if something is wrong with the computer itself.

Ok your explanations makes things very clear now, thank you for your help :)

As the new versions coming out are mainly making improvements on GPU calculations which I'm not using for the moment I'll stay on the old version that gives maximum of points!! (even if it might be doing less data crunching it's sad to admit it^^)