Could someone please explain the very different results I got from Boinc -run_cpu_benchmark? These are from the same dual-boot PC. Benchmark got nearly 100% of processor time in each test.

Linux (SuSe 9.1 64 bit)
Measured floating point speed: 1072.97 million ops/sec
Measured integer speed: 2645.31 million ops/sec
Measured memory bandwidth: 953.67 MB/sec

Windows XP Home
Measured floating point speed: 2469.62 million ops/sec
Measured integer speed: 5072.19 million ops/sec
Measured memory bandwidth: 953.67 MB/sec

Hello Juha,

I also noticed that (tought I do not have windows on my PC).
Can someone explain this please ?

this is the Spirit of "NOVELL" LINUX i THINK ;-)))
[/url] [/url]


IN MEMORY OF CARL SAGAN.1934-1996

Funny ;-)))

But I do not run Suse/Novell. I run a self made distro (Linux From Scratch) ...
Maybe it's the way the boinc program is compiled or maybe the option I choosed
for my kernel compilation (tought I compiled it for P4 CPU)...

> Could someone please explain the very different results I got from Boinc
> -run_cpu_benchmark? These are from the same dual-boot PC. Benchmark got
> nearly 100% of processor time in each test.
>
> Linux (SuSe 9.1 64 bit)
> Measured floating point speed: 1072.97 million ops/sec
> Measured integer speed: 2645.31 million ops/sec
> Measured memory bandwidth: 953.67 MB/sec
>
> Windows XP Home
> Measured floating point speed: 2469.62 million ops/sec
> Measured integer speed: 5072.19 million ops/sec
> Measured memory bandwidth: 953.67 MB/sec
>
>
64 bit operating systems are not faster cuase they have higher
overheads. 64 bit doesn't mean it's faster, it only allows applications
to use more then 4GB of ram. Since BOINC or SETI only uses 16MB of ram,
they won't benifit at all from 64 bit. In fact as you can see, they lose out.

> 64 bit operating systems are not faster cuase they have higher
> overheads. 64 bit doesn't mean it's faster, it only allows applications
> to use more then 4GB of ram. Since BOINC or SETI only uses 16MB of ram,
> they won't benifit at all from 64 bit. In fact as you can see, they lose out.
>
>
I don't think it's linked to the 64 bits feature. I run a linux kernel
2.4.25-SMP (on a uniprocessor machine - P4 1,7 GHz without HT) and compared
to other processor of the same class but with Windows I have +/- the half of
the "performance".

I have the feeling that this is more linked to how the program or the OS is
compiled (with or without omptimization).

I don't know what could explain this difference. Maybe should I try to compile
a version of seti and see if there is any difference.

The linux benchmark was fine in 3.07 but they made some changes between there and 3.18. They know about it but I have yet to see anyone really post a response.

> The linux benchmark was fine in 3.07 but they made some changes between there
> and 3.18. They know about it but I have yet to see anyone really post a
> response.
>
>

I think they are very busy with the set-up of new "split processes" (I don't know how they call this process but it's the processes that prepare the WU). There is a shortage of WU and I understand that their priorities are to resolve this shortage.

> > The linux benchmark was fine in 3.07 but they made some changes between
> there
> > and 3.18. They know about it but I have yet to see anyone really post a
> > response.
> >
> >
>
> I think they are very busy with the set-up of new "split processes" (I don't
> know how they call this process but it's the processes that prepare the WU).
> There is a shortage of WU and I understand that their priorities are to
> resolve this shortage.
>
>

Makes perfect sense to me. THey have to get the priorities take care of first but I am sure it is in the queue.

Hi there.

Hope it is ok, that i post a question here. is it about Linux OS, is it better than microsoft OS to crunch? i've heard some people saying that Linux uses resources much better than Microsoft OS. Have thought of trying Linux

thanks
<font face="comic sans ms">THE CRUNCHER'S CORNER !!!</font>

All my benchmarks on linux are now very different than they were in beta. However, my systems are still running the wu's in the same amount of time and asking for credits in the same range as they did in beta.

There were many problems in beta with linux and windows systems asking for vastly different amounts of credit - often linux systems would ask for something in the range of 30 points, then a windows system would come along and ask for something in the range of 4 or 5 points. Your benchmarks are a very large factor in determining how much credit you ask for.

A lot of the difference could be accounted for by the fact that the windows version had some debug code still in it, while there was no debug code in the linux version - but even after the debug was removed from windows, the credit requests still never matched up.

Though I haven't seen anything indicating what was changed, it appears that they solved that problem by making some changes to the way the benchmarks are run - as linux and windows systems are now requesting very similar points for a wu, even though they're reported very different benchmarks.

If the choice is have the same benchmarks and different credit requests or different benchmarks and the same credit requests, well, I can live with the different benchmarks.

I've noticed the same thing. Initially I thought that it was because I was using a dual-processor machine, and linux was giving the benchmark per CPU, while Windows was reporting the total for both (which would explain why the Windows benchmarks are just over twice as fast), but I guess that's not the case.

I don't think it's a compiler issue as some have suggested - I mean, over a factor of two?

> Though I haven't seen anything indicating what was changed, it appears that
> they solved that problem by making some changes to the way the benchmarks
> are run - as linux and windows systems are now requesting very similar
> points for a wu, even though they're reported very different benchmarks.

If that's the case, that would be great. But I was under the impression that credit claimed was essentially given by the benchmark numbers multiplied by the time taken to process a WU. If Windows benchmarks are twice as high as Linux ones, but both OSes take roughly the same time to process a WU, wouldn't a Windows system clock up twice as much credit per WU?

Any clarification gratefully accepted!

Mattias