My PC(AMD 64 3200+ with 512 MB DDR 400 Mhz) is pretty slow with the new boinc client. With the old client I had around 2 s/TS and the new boinc client should bring 30% performance. But now I have a performance of 2,3 s/TS. I can\'t agree with this. that\'s to slow for that CPU!!

Founder of

<a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/team_display.php?teamid=36"><img src="http://www.ug-abi.de/2008/projekt1.jpg"></a>

I believe there is a performance drop for AMD and a boost for intel :(

&gt; I believe there is a performance drop for AMD and a boost for intel :(
&gt;
I haven't checke with my AMD64 but yes, P4 are singificantly faster: one of mine was doing 2.4 with classic and now about 1.8 (!?)

I didn't lose as much as you Admiral Hawkes, but went from ~2.0 on CPDN classic to 2.08 BOINC CPDN Windows and 2.11 BOINC CPDN Linux. Disappointing to have a slowdown with the new client. Didn't expect it at all.

Have you all considered the little points?

1) There is no reliable 64bit Windows yet to help your 64bit CPU.
2) There is no BOINC yet that will reliably crunch on 64bit OSes (it will revert back to 32bit) &gt;&gt; This in case you're running a 64bit Linux.
3) All AMDs are only running at a PR rated speed towards the Pentiums. You may want to expect to have it run at 3.2Ghz, but it isn't. The benchmarks won't change a bit about it.
4) I wouldn't complain about 0.3 s/TS on a beta project. ;)
5) CPDN on BOINC likes lots of memory. You may want to test it again with 1GB or 2GB of PC3200 installed.
--------------------
Jordâ„¢

&gt; Have you all considered the little points?
&gt;
&gt; 1) There is no reliable 64bit Windows yet to help your 64bit CPU.
&gt; 2) There is no BOINC yet that will reliably crunch on 64bit OSes (it will
&gt; revert back to 32bit) &gt;&gt; This in case you're running a 64bit Linux.
&gt; 3) All AMDs are only running at a PR rated speed towards the Pentiums. You may
&gt; want to expect to have it run at 3.2Ghz, but it isn't. The benchmarks won't
&gt; change a bit about it.
&gt; 4) I wouldn't complain about 0.3 s/TS on a beta project. ;)
&gt; 5) CPDN on BOINC likes lots of memory. You may want to test it again with 1GB
&gt; or 2GB of PC3200 installed.
&gt; --------------------
&gt; Jordâ„¢
&gt;

Hi Jord,
i think that 1-3 apply also to classic as well.
ad for 5) this is the same for 1 model, takes more memory when two models at once (some P4 HTs users do so).
About 4) - well, the client has alerady some improvments (like some file i/o optimalization -&gt; less HD intensive). Client core will propably not change much (some memory issue perhaps like it was on MacOs) and what will/should change is 50% CPU with visualization and perhaps some client-server communication.
To be brief, i don't expect significant performance changes when CPDN/BOINC goes gold (public version).

Mikey just posted a link where Carl prepared some BOINC CPUs performance table.
http://www.climateprediction.net/board/viewtopic.php?p=18547#18547

Direct link:
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/cpu.html

&gt; To be brief, i don't expect significant performance changes when CPDN/BOINC
&gt; goes gold (public version).
&gt;
That's why it's in beta. :)

To find quirks like this. But okay, I never ran classic long enough to see the difference for points 1 to 3 &amp; 5. Or changed my hardware inbetween. Although, in saying that, I did run classic for over 48 hours and afaik never got credit for it (trickle).

What I did find though is that this BOINC client is better for the P4 Celerons. For once I am not requesting huge amounts of credit that I may never get.

As for point 3 in my list above, the PR rating is set against a gaming CPU. As far as I know, the AMDs aren't that good in 3D CAD/CAM if you compare them in raw speed against the P4s they are competing against.

But then again, we all have problems running the OpenGL graphics. ;)

P.S: Mikey is a bit slow, the link's been in http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=67#397 since yesterday. ;)
--------------------
Jordâ„¢

&gt; Have you all considered the little points?
&gt;
&gt; 1) There is no reliable 64bit Windows yet to help your 64bit CPU.
&gt; 2) There is no BOINC yet that will reliably crunch on 64bit OSes (it will
&gt; revert back to 32bit) &gt;&gt; This in case you're running a 64bit Linux.
&gt; 3) All AMDs are only running at a PR rated speed towards the Pentiums. You may
&gt; want to expect to have it run at 3.2Ghz, but it isn't. The benchmarks won't
&gt; change a bit about it.
&gt; 4) I wouldn't complain about 0.3 s/TS on a beta project. ;)
&gt; 5) CPDN on BOINC likes lots of memory. You may want to test it again with 1GB
&gt; or 2GB of PC3200 installed.
&gt; --------------------
&gt; Jordâ„¢
&gt;
None of this applies to the concerns raised by the A64 owners. The concern was about the BOINC client running slower on A64s than the classic client. Speculation is that it's due to the Intel Fortran compiler now being used. While it is great that the Intel CPUs are running faster (more model years crunched in a given period of time by those CPUs, good for the project), I hate to see those with AMD CPUs (me included) crunch less years in a given period of time. BTW, AMD CPUs are preferred for most scientific apps because of their floating point performance and now, the extremely low latency of the A64s memory controller.

&gt; P.S: Mikey is a bit slow, the link's been in
&gt; http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=67#397 since
&gt; yesterday. ;)
&gt; --------------------
&gt; Jordâ„¢
Hi Jordâ„¢,

How well you know me! I just can’t keep up with the two User Forums, the original CPDN and now this BOINC orientated one....

FYI there was a benchmark summary for the original CPDN that has been summarized in the following CPDN links:

http://www.climateprediction.net/board/viewtopic.php?t=2018
http://cpdn.tuxie.org/CPU_Benchmark/CPDN_CPU_Benchmark_Summary.xls
http://cpdn.tuxie.org/CPU_Benchmark/CPDN_CPU_Benchmark_Summary.pdf

UK4CP @ www.uk4cp.co.uk (United Kingdom Group) Celeron 2.6GHz XP Pro SP1 768MB RAM

geophi -- I'm going to look into AMD64's, not least because that's what I have at home! We had to move from the Compaq Fortran compiler for Win since it doesn't run in Visual Studio.NET which is needed for BOINC, and perhaps Compaq was slightly "friendlier" to AMD's than the Intel compiler was. A lot of it is economics, we can use our old licence for Intel Fortran on Linux so it made sense to do that for Intel on Windows etc. We just don't have the cash for buying a lot of different compilers.

If anyone is interested, the Intel Fortran Windows compiler settings I use are:

"Favor Fast Code"
"Optimize for Intel Processor: Pentium Pro, Pentium II, Pentium III"
"Use Intel Processor Extensions: Intel Pentium III and compatible"
"Require Intel Processor Extensions: None"

So this supposedly generates Pentium-optimized code as well as "generic IA-32 code" (from the help msg). So I would have thought AMD's were at least compatible at the P &amp; P2 levels, but perhaps not. Or it's that boost of P3s &amp; P4s that is putting Intel Pentium's ahead? Anyway, it's similar to my old Compaq Fortran settings, I believe they were "Pentium (I)" optimized instructions since it was assumed at least a first-gen Pentium would be running this!

In Linux, I just use the "-O3" optimizations since anything higher seems to cause instabilities in the climate model.

I must also admit that it is not easy to cope with both forum and read all messages - i'm at home for 2 hours/day now and spending 1 hours with e-mails and CPDN - not enought.

Actually it was me who overlooked Carl's link (wasn't fast enough as Jord would say, and wanted to start BOINC benchmarks on already mentioned classic CPDN benchmark on classic forum...

&gt; &gt; P.S: Mikey is a bit slow, the link's been in
&gt; &gt; http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=67#397
&gt; since
&gt; &gt; yesterday. ;)
&gt; &gt; --------------------
&gt; &gt; Jordâ„¢
&gt; Hi Jordâ„¢,
&gt;
&gt; How well you know me! I just can’t keep up with the two User Forums, the
&gt; original CPDN and now this BOINC orientated one....
&gt;
&gt; FYI there was a benchmark summary for the original CPDN that has been
&gt; summarized in the following CPDN links:
&gt;
&gt; http://www.climateprediction.net/board/viewtopic.php?t=2018
&gt; http://cpdn.tuxie.org/CPU_Benchmark/CPDN_CPU_Benchmark_Summary.xls
&gt; http://cpdn.tuxie.org/CPU_Benchmark/CPDN_CPU_Benchmark_Summary.pdf
&gt;
&gt; UK4CP @ www.uk4cp.co.uk (United Kingdom Group) Celeron 2.6GHz XP Pro SP1 768MB
&gt; RAM
&gt;

@carl:
have you tried -tpp7, which imho uses sse2 on p4, athlon64...?
or does the model not like this?

&gt; @carl:
&gt; have you tried -tpp7, which imho uses sse2 on p4, athlon64...?
&gt; or does the model not like this?

I can't remember if I tried that, I may give it a go and see if it works OK.

&gt; &gt; @carl:
&gt; &gt; have you tried -tpp7, which imho uses sse2 on p4, athlon64...?
&gt; &gt; or does the model not like this?
&gt;
&gt; I can't remember if I tried that, I may give it a go and see if it works OK.
&gt;
&gt;

I once more read the ifort manual, now i think -tpp7 doesn't use see2, it only optimizes for p4s. to use sse2 you would need something like -axN

&gt; &gt; @carl:
&gt; &gt; have you tried -tpp7, which imho uses sse2 on p4, athlon64...?
&gt; &gt; or does the model not like this?
&gt;
&gt; I can't remember if I tried that, I may give it a go and see if it works OK.
&gt;
It seems to me that those different switches will results in having different application version for different CPU:
- default for older CPU,
- extra complitaliton for P4,
- extra compilation for AMD64.

Now that BOINC can better identify CPUs, this should be possible.
I'm also curious how AMD64 would do with proper compilation...

btw, i moved BOINC client to different machine (both are P4) to let it continue there - no problems so far, now in Phase 2.

A few more links regarding the Intel Fortran Compiler for Windows/Linux. There seems to be good advice for Pentium, of course, as most information comes from Intel! Somebody also asked about AMD processors back in May 2004, but no answer as of yet. It looks like that maybe optimal performance for all environments, whether software or hardware (I.E. Windows, Linux, Intel/AMD processors) might require some compilation tuning.

http://www.lrz-muenchen.de/services/compute/linux-cluster/develop/#publish2.3.0.0.0.0
http://support.intel.com/support/performancetools/fortran/sb/CS-008948.htm
http://support.intel.com/support/performancetools/fortran/performOpt.htm
http://www.dbforums.com/t914914.html

Of course AMD don’t really consider the Intel Fortran Compiler suite of software as “best” for their processors:

http://www.amd.com/us-en/assets/content_type/Additional/30201B_-_AMD64_Tools_Guide_All_OS.pdf

UK4CP @ www.uk4cp.co.uk (United Kingdom Group) Celeron 2.6GHz XP Pro SP1 768MB RAM

hi,

i'm running P4-2,533GHz (512Mb DDR333 cl2,5, WinXP) with 2,60TS/s ... that ist quite small, i think.

now i want to compare with the original CPDN-software. the stats for this machine from the statspage:
Machine Name HAL9000
Total Model Years 135.024
Total CPU Days 89.422
Total Petacycles 19.325
Completed Full Runs 3
Completed Short Runs 0

how to compute avg. TS/s from this variables?

thanks

&gt; hi,
&gt;
&gt; i'm running P4-2,533GHz (512Mb DDR333 cl2,5, WinXP) with 2,60TS/s ... that ist
&gt; quite small, i think.&gt;
Hi gandhi,

Hmmm, OK there are 259,248 timesteps per phase, 3 phases per model and 45.008 years per model. So we can presume HAL9000 completed 3 models in 89.422 CPU days. There are 86,400 (60*60*24) CPU seconds in one day, so HAL9000 completed 3 models in ~7,726,060.8 CPU seconds. HAL9000 completed 3 models, 9 phases and therefore 2,333,232 timesteps. So, ~3.3 seconds per timestamp.

Clearly for BOINC like the rest of Pentium 4 users you are benefiting from the ~25-30% performance increase.

To save my fingers and my poor math maybe you could study the following links for other CPDN classic and BOINC performance figures:

http://climateapps2.oucs.ox.ac.uk/cpdnboinc/cpu.html
http://cpdn.tuxie.org/CPU_Benchmark/CPDN_CPU_Benchmark_Summary.xls
http://cpdn.tuxie.org/CPU_Benchmark/CPDN_CPU_Benchmark_Summary.pdf

<i> UK4CP @ www.uk4cp.co.uk (United Kingdom Group) Celeron 2.6GHz XP Pro SP1 768MB RAM<i>

although with "classic CPDN" that is "wall clock time" of 3.3 seconds per timestep, whereas BOINC is actual CPU time. So if your CPU was used a lot for things other than cpdn the old cpdn times are underestimating.

I tried the -tpp7 switch today and didn't notice any big improvement. When I get time I will investigate other Fortran compilers. I haven't found any AMD64 compiler yet that works with the model; it's a very touchy million lines!