[mikkyo]

Posted on the team resources page here:
Team MacNN Boinc Clients

[jeffreyloaf]

Well I installed G5 superbench for my dual 2.5. With processor performance set to highest; my results are as follows.
4567 double precision MIPS (Whetstone) per CPU
16817 integer MIPS (Dhrystone) per CPU

I am using boinc manager and it tells me it takes 36 and half minutes per work unit.

Good job mikkyo

[Lateralus]

2005-06-08 22:11:21 [---] 3647 double precision MIPS (Whetstone) per CPU
2005-06-08 22:11:21 [---] 13444 integer MIPS (Dhrystone) per CPU

iMac G5 2GHz.

Thanks Mikkyo.

[beadman]

Thanks, mikkyo!!
My 1.3GHz iBook went from

Measured floating point speed 684.26 million ops/sec
Measured integer speed 1605.37 million ops/sec

to

1963 double precision MIPS (Whetstone) per CPU
5547 integer MIPS (Dhrystone) per CPU

I'm running the G4 Superbench version in Menubar. Started right up, ran the bench marks, and resumed computation on the WU it had been working on before.

beadman

[Shaktai]

There is a known glitch on some machines, if you are running the "Advanced GUI" with BOINC Manager. (I don't think it impacts the menubar GUI).

The problem is that BOINC Manager, can't communicate, right after the upgrade. The fix is to "restart" your computer. To the best of my knowledge, this only affects the BOINC Manager. The CLI works just fine without problems, and the menubar version should to, but I haven't confirmed it.

[electrician1972]

What is the advantage (if any) to these 4.44 superbench clients? I setup my system with the G4 "superbench" version and the benchmarks reported back are 1574 Whetstone and 4418 Dhrystone. About a 3x increase from previous versions. Is this a more accurate representation of the processing power of the PPC chips vs. Intel chips ?

[mikkyo]

The superbench clients are good for farmers running dedicated systems that crunch all the time as they will download a more accurate daily amount of work and keep the machine quite busy.
Anyone can use them, however they could negatively effect your production on your main project.

If you are attached to several projects and also use the machine frequently, or sleep it some of the time, the superbench clients will result in the machine probably downloading more work than it can possibly complete before the due dates.
With boinc_4.44's project based priority scheduling, you may still get much of that extra work done in time for credit, however you won't spend as much time crunching on your project of choice as you might have or might like to have, as your over-utilized machine will be having to process those extra work units for the other projects to avoid overshooting the deadline.
There are some fail-safes in boinc_4.44 though, and it will prevent you from getting new work if you are over-committed while it crunches imminently due projects - for example, it will stop working on the main project, say Predictor, in order to complete due soon workunits for ClimatePrediction and Einstein.

So in short, superbench gives you more workunits per day and works best for dedicated machines or few attached projects.
The regular superoptimized boinc_4.44 clients will be better in the long run for slower machines, or with down time or heavy use and/or machines attached to many projects.

As for the power of PPC vs Intel chips, their is little difference as long as you have work - both processor types will crunch to their max ability with an optimized boinc client running.
Now if you have an optimized worker, like the altivec seti worker, on a dedicated machine, the superbench much more accurately represents the power of your machine.
It is also a very good choice if you run in Console mode with most resources disabled, have gobs of RAM, and run only the boinc client at nice of -20 on a dedicated machine.