Hi, I just installed mandrake 10.1, and I run the command line version of boinc.

This machine used to run boinc_gui under windows xp before.

The benchmark under windows xp was 877.47 for fp and 2139.93 for integer.

Now under linux it's only 520.96 fp and 1096.72 integer.

Why is the linux client slower? Under windows it was using 99% cpu, and when I see the boinc process in linux it also uses about 97-98% cpu.

This is on a PIII 1GHZ.

Thanks!

---

The windows compiler was optimizing out the benchmarks. This should be fixed in 4.25.

---

BOINC WIKI

BOINCing since 2002/12/8

> Why is the linux client slower?

There are two problems and three possible solutions. The problems:

1. The old Windows clients "cheated" at the benchmarks.
2. The Berkeley-distributed Linux clients are not properly optimised.

The possible solutions are:

1. Compile your own Boinc client. There are instructions here.
2. That same site has pre-compiled optimised clients - although only up to version 4.19 at the moment.
3. Run the Windows client in Wine.

---

Be lucky

Neil

I have the optimized version (Using mandrakelinux too!) and its average credit is higher than my windows after changing it. Just check my PCs

---

> > Why is the linux client slower?
>
> There are two problems and three possible solutions. The problems:
>
> 1. The old Windows clients "cheated" at the benchmarks.
> 2. The Berkeley-distributed Linux clients are not properly optimised.
>
> The possible solutions are:
>
> 1. Compile your own Boinc client. There are instructions <A> HREF="http://www.pperry.f2s.com/boinc-compile.htm"> here[/url].
> 2. That same site has pre-compiled optimised clients - although only up to
> version 4.19 at the moment.
> 3. Run the Windows client in Wine.
>
>
>

Thanks a lot! I just compiled the client and I'm now at 988 fp and 1547 integer. I will try to play with the compile options (I only tried march=pentium3 and O3), but I'm already a lot better than before.

Thanks again!

---

> Thanks a lot!

Glad to help.

> I just compiled the client and I'm now at 988 fp and 1547
> integer. I will try to play with the compile options (I only tried
> march=pentium3 and O3), but I'm already a lot better than before.

Those plus -fomit-frame-pointer -funroll-loops -ffast-math -fforce-addr -ftracer are what work for most of us. ;)

Bear in mind, though, that this optimisation does not enable you to process WUs any faster - it just increases your (and other peoples') credit sometimes. ;)

---

Be lucky

Neil