Performance Differences

[Dyyryath]

I've noticed that the Windows client is quite a bit faster than the Linux one. :(

I've got two virtually identical systems here (A64-3500, 2gb RAM):

Windows Avg Credit: 352.66
Linux Avg Credit: 270.49


Since this appears to be the case, I'm now running the Windows client on the Linux machine under VMWare which, hopefully, will bring its numbers up some. Has anyone else tried this tactic? Has anyone tried running BOINC under Wine?

[Bok]

Is this using the optimized clients ?

[Dyyryath]

Yeah, the 5.x windows client listed in your Optimized Clients sticky thread and the optimized 4.72 client you sent me the other day.

[Dyyryath]

I'm not sure exactly how the average credit is calculated (or whether it's a good metric), but an A64-3200 I just started running the Windows client on under VMWare has gone from 266 to 330.

[Bok]

What do the benchmark scores say ?

The RAC is calculated over a number of days (or even weeks), so it is not entirely accurate at first.

Bok

[LAURENU2]

not so loud it is supose to be a secret:Pokes: :harhar:

[Dyyryath]

LAURENU2: Sorry bout that. :D

Bok: On the A64-3200 box:

Native Linux: 2120/4064
VMWare Windows XP: 1651/6102


Which does BOINC rely on anyway, integer or floating point? I notice the floating point is better under Linux, but running in Windows under VMWare puts the integer bench a pretty good way over the top...

[Bok]

I think it's the integer, though that is a huge score for a single A64, considering I get 9000 on an x2 4200+ @ 2.5 under win Xp with the optimized client.

Bok

[MerePeer]

I grabbed the 5.2.6 source out of cvs and solved all the linux prereq installs. Now I'm stuck with some weird sed error during ./configure from a script that is generated during configure. Maybe it is looking for gsed and maybe my (debian) sed isn't quite the same?

checking for pthread_join in dynamic library pthread... -lpthread
configure: creating ./config.status
config.status: creating ./version.h
sed: file ./confstatxZ9St6/subs-2.sed line 32: unterminated `s' command
config.status: creating api/Makefile
sed: file ./confstatxZ9St6/subs-2.sed line 32: unterminated `s' command
config.status: creating apps/Makefile
sed: file ./confstatxZ9St6/subs-2.sed line 32: unterminated `s' command
config.status: creating clientgui/Makefile
sed: file ./confstatxZ9St6/subs-2.sed line 32: unterminated `s' command
config.status: creating client/Makefile
sed: file ./confstatxZ9St6/subs-2.sed line 32: unterminated `s' command
config.status: creating client/win/boinc_path_config.py
sed: file ./confstatxZ9St6/subs-2.sed line 32: unterminated `s' command
...

[Bok]

That looks very strange. Configures and compiles fine on gentoo.

You are doing

./configure --disable-server right??

Bok

[MerePeer]

Yes I was doing ./configure --disable-server.

I dropped 5.2.6, grabbed 5.2.7 and it built perfectly. Debian etch.

I've optimized by adding the arch-type and O3 compiler opts only and so far it seems the AMD arch are getting about a 10% Integer calc increase and the Pentium arch havent changed. Need to run these a few days to see if they hold up.
;)