Hm, but the differences are very very big and it shouldn't be like that... If it goes on like this you just feel like wasting very much CPU Time, because the Windows program seems to have bugs or whatever...
You don't waste the CPU time because each results are important.
So, this is not the reason of your feeling.
It is, as I will constantly get the feeling that my CPU can do more science work in the same amount of time on another project.
Hm, but the differences are very very big and it shouldn't be like that... If it goes on like this you just feel like wasting very much CPU Time, because the Windows program seems to have bugs or whatever...
You don't waste the CPU time because each results are important.
So, this is not the reason of your feeling.
It is, as I will constantly get the feeling that my CPU can do more science work in the same amount of time on another project.
How did you compare the science works between different projects?
What is the gravitation / protein ratio?
Yep, really strange. Intel seems to be affected, too, in some cases. Akos, would you mind having a look at the "Information about S5" thread? We have some experiments running there about CPU performance, especially on Intels.
What was the question in that thread? I didn't see.
My Core2 runs an SSE optimised version of XLALComputeFaFb subroutine. It shows about 70% performance improvement. Bernd will implement it into the source code and compile for all x86 based platforms.
Yep, really strange. Intel seems to be affected, too, in some cases. Akos, would you mind having a look at the "Information about S5" thread? We have some experiments running there about CPU performance, especially on Intels.
What was the question in that thread? I didn't see.
My Core2 runs an SSE optimised version of XLALComputeFaFb subroutine. It shows about 70% performance improvement. Bernd will implement it into the source code and compile for all x86 based platforms.
A question arising from the experiments is whether it's true that systems not supporting SSE2 will get a heavy performance penalty in the Win app because a fallback version of modf is used my the math lib, and this non-SSE modf implementation seems to be quite slow compared to that in the linux app.
Could anything be done about this in the next version? This would affect not only the slower Pentium IIIs but also the many AMD Athlon XPs that lack SSE2, but would otherwis benefit from your upcoming SSE optimization.
The other question from that thread is why there is a slowdown under Win for some systems (even those that do support SSE2), and not for others, even if they are very similar (it's not just the AMD boxes).
P.S.: Thanks in advance for the SSE optimization you mentioned, I'm looking forward to it.
P.S.: Thanks in advance for the SSE optimization you mentioned, I'm looking forward to it.
Agreed, and I'm sure a friend of mine would be very pleased =)
Anyway, currently the Win version is slower on some platforms, but the Linux boxes had stability problems until recently (wehn the 4.21 app was released).
This is the first BOINC project on gravitational wave detection, so in a way we all go where noone else has gone before...Its only natural that some problems for this or that OS / compiler / CPU will surface...and be solved. Evolution of the software is part of the experiment and to me it's half the fun to see this happen.
Been away for a while - we have power issues down here in Cape Town...
I have finally finished a WU in Linux and am now going for one in Win XP. They are being done on a dual boot system, and my usuage on each side is similar. Granted a single WU hardly is a pattern, but **might** be an indication
Hardware: Pentium 4 CPU 2.66GHz [x86 Family 15 Model 2 Stepping 9] with 768 megs RAM. No overclocking, tho the Linux is Gentoo, which is supposed to be optimal in that it is compiled for the hardware, and the WinXP has been tweaked to within an inch of its life :) .
The WU are looking to be similar in length, with possibly Linux having the edge - 42 hours actual time, and Win XP estimates 44 hours.
From what I have read the Linux/WinXP difference is less with Intels ?
Anyhow, I'll finish this info in around 40 - 50 hours - sigh...
RE: RE: Hm, but the
)
It is, as I will constantly get the feeling that my CPU can do more science work in the same amount of time on another project.
RE: RE: RE: Hm, but the
)
How did you compare the science works between different projects?
What is the gravitation / protein ratio?
RE: How did you compare the
)
I meant science in general, hence the reason I didn't state what type of science.
Hi Annika! RE: Yep,
)
Hi Annika!
What was the question in that thread? I didn't see.
My Core2 runs an SSE optimised version of XLALComputeFaFb subroutine. It shows about 70% performance improvement. Bernd will implement it into the source code and compile for all x86 based platforms.
RE: Hi Annika!RE: Yep,
)
A question arising from the experiments is whether it's true that systems not supporting SSE2 will get a heavy performance penalty in the Win app because a fallback version of modf is used my the math lib, and this non-SSE modf implementation seems to be quite slow compared to that in the linux app.
Could anything be done about this in the next version? This would affect not only the slower Pentium IIIs but also the many AMD Athlon XPs that lack SSE2, but would otherwis benefit from your upcoming SSE optimization.
The other question from that thread is why there is a slowdown under Win for some systems (even those that do support SSE2), and not for others, even if they are very similar (it's not just the AMD boxes).
P.S.: Thanks in advance for the SSE optimization you mentioned, I'm looking forward to it.
CU
BRM
RE: P.S.: Thanks in advance
)
Agreed, and I'm sure a friend of mine would be very pleased =)
RE: RE: P.S.: Thanks in
)
Anyway, currently the Win version is slower on some platforms, but the Linux boxes had stability problems until recently (wehn the 4.21 app was released).
This is the first BOINC project on gravitational wave detection, so in a way we all go where noone else has gone before...Its only natural that some problems for this or that OS / compiler / CPU will surface...and be solved. Evolution of the software is part of the experiment and to me it's half the fun to see this happen.
CU
BRM
Hello Been away for a
)
Hello
Been away for a while - we have power issues down here in Cape Town...
I have finally finished a WU in Linux and am now going for one in Win XP. They are being done on a dual boot system, and my usuage on each side is similar. Granted a single WU hardly is a pattern, but **might** be an indication
Hardware: Pentium 4 CPU 2.66GHz [x86 Family 15 Model 2 Stepping 9] with 768 megs RAM. No overclocking, tho the Linux is Gentoo, which is supposed to be optimal in that it is compiled for the hardware, and the WinXP has been tweaked to within an inch of its life :) .
The WU are looking to be similar in length, with possibly Linux having the edge - 42 hours actual time, and Win XP estimates 44 hours.
From what I have read the Linux/WinXP difference is less with Intels ?
Anyhow, I'll finish this info in around 40 - 50 hours - sigh...
Gray
Not with every Intel
)
Not with every Intel unfortunately, we had a Core and a few P3s who got hit rather badly...
RE: Not with every Intel
)
Hi Annika
Just by way of interest, could you summarise WU times relative to cpu power for any pcs you have near in hardware specs to mine own ?
Gray