climateprediction.net home page
Client performance

Client performance

Message boards : Number crunching : Client performance
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 . . . 6 · Next

AuthorMessage
Christian Hoklas

Send message
Joined: 5 Aug 04
Posts: 11
Credit: 63,408
RAC: 0
Message 356 - Posted: 7 Aug 2004, 22:28:13 UTC

My iBook G4 / 933 needs 9.53 secs per timestep. This is fairly slow compared to the average speed reported here by pc owners. Maybe there is somebody with a G5 out there, telling us what he/she gets out of the client?! Are there any other mac users out there at all? I am feeling so lonely... :-))

Thanks to the climateprediction.net team for bringing cpdn to my favourite platform. ;-)
ID: 356 · Report as offensive     Reply Quote
old_user1
Avatar

Send message
Joined: 5 Aug 04
Posts: 907
Credit: 299,864
RAC: 0
Message 361 - Posted: 7 Aug 2004, 23:38:13 UTC - in response to Message 356.  

> My iBook G4 / 933 needs 9.53 secs per timestep. This is fairly slow compared
> to the average speed reported here by pc owners. Maybe there is somebody with
> a G5 out there, telling us what he/she gets out of the client?! Are there any
> other mac users out there at all? I am feeling so lonely... :-))
>
> Thanks to the climateprediction.net team for bringing cpdn to my favourite
> platform. ;-)

Unfortunately that's normal speed for G4's, the Fortran compilers just aren't as spiffy as what Intel does for Win & Linux boxes. We had a "parallelized" CPDN that took advantage of the Mac's altivec processing but unfortunately the computations weren't correct (there's a million lines of code in the model so it's tough to change/optimize code).

I've seen G5 stats from the alpha test and a G5 is getting around P4/2.8GHz levels (i.e. 2.5-2.8 seconds per timestep). Our only Mac here (our development & test box -- takes about two hours just to compile the climate model!) gets 10.5 seconds per timestep (it's a G4/800MHz iBook).

ID: 361 · Report as offensive     Reply Quote
old_user357

Send message
Joined: 7 Aug 04
Posts: 7
Credit: 14,787
RAC: 0
Message 362 - Posted: 7 Aug 2004, 23:43:21 UTC - in response to Message 361.  

> Unfortunately that's normal speed for G4's, the Fortran compilers just aren't
> as spiffy as what Intel does for Win & Linux boxes. We had a

And I thought FORTRAN was a dead language... ;)
ID: 362 · Report as offensive     Reply Quote
old_user1
Avatar

Send message
Joined: 5 Aug 04
Posts: 907
Credit: 299,864
RAC: 0
Message 363 - Posted: 7 Aug 2004, 23:49:41 UTC - in response to Message 362.  

> And I thought FORTRAN was a dead language... ;)

I thought so too until I joined this project, but there's billions of lines of code out there still happily running, especially in the science field I guess. It would take us 20 years to convert it to C! :-)

ID: 363 · Report as offensive     Reply Quote
old_user355

Send message
Joined: 7 Aug 04
Posts: 187
Credit: 44,163
RAC: 0
Message 365 - Posted: 8 Aug 2004, 0:37:53 UTC

I can't get the visualizations to work, but my trickles page shows three trickle results. Two 1.76, and a 1.77.

AMD FX-53, Win XP Pro SP1.
ID: 365 · Report as offensive     Reply Quote
Profile astroWX
Volunteer moderator

Send message
Joined: 5 Aug 04
Posts: 1496
Credit: 95,522,203
RAC: 0
Message 370 - Posted: 8 Aug 2004, 1:26:14 UTC
Last modified: 8 Aug 2004, 1:26:49 UTC

HT P4s running SuSE Linux 9.0
Abox 2.8 GHz, two models @ 3.8 sec/TS each
Bbox 3.0 GHz, two models @ 3.7 & 3.8 sec/TS

AVG. throughput ~1.9 sec/TS for both boxes (if that is the correct way to evaluate HT performance).

Values were calculated from a full Trickle per run (11+ hours & 10802 TS each). (My 'AVG=' numbers are invalid because the four runs were carried forward from Alpha and the Alpha CPU time was lost in the process.)

The odd thing is that the 3.0 box has an 8MB Cache IDE HD (theoretically a bit faster than the older flavor); the 2.8 has the older 4.0 Meg cache IDE HD.
The 3.0 box has 2.0 latency memory, the 2.8 box has 'off the shelf' (presumably 2.5). Both machines have 1.0 gig, dual channel.
(The faster box does browsing and email -- but 'gaming' for me is an occasional hand of Patience. Shouldn't zap all that many cycles...? [Either the 3.0 is a wee bit of a dog, or I have an energetic 2.8.])
________________________________________________
We have met the enemy and he is us -- Pogo
ID: 370 · Report as offensive     Reply Quote
old_user194

Send message
Joined: 5 Aug 04
Posts: 63
Credit: 21,399,117
RAC: 0
Message 383 - Posted: 8 Aug 2004, 4:42:00 UTC

Christian>My iBook G4 / 933...
Ha! Ha! I beat you! I have one PC doing 13.4s/TS. But then its a Celeron 1100...

Cheers,

PeterV.
ID: 383 · Report as offensive     Reply Quote
old_user56

Send message
Joined: 5 Aug 04
Posts: 4
Credit: 33,247
RAC: 0
Message 393 - Posted: 8 Aug 2004, 9:26:41 UTC

P4 2,8 Ghz: 3,6s/Ts running 2 Boinc modells
Centrino 1,4Ghz : 2,66 s/TS running 1 Boinc model
ID: 393 · Report as offensive     Reply Quote
old_user1
Avatar

Send message
Joined: 5 Aug 04
Posts: 907
Credit: 299,864
RAC: 0
Message 397 - Posted: 8 Aug 2004, 9:42:17 UTC - in response to Message 393.  
Last modified: 8 Aug 2004, 9:44:08 UTC

I have made a prelimary CPU-time/comparison page (based on 1000 trickles) here:

<A HREF="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/cpu.html">CPU Timings So Far</A>

I think I was a bit off in my original "spam beta invitation" email, a P4 2.6GHz looks like it would complete in just under 3 weeks, not 4 weeks (running "24/7")
ID: 397 · Report as offensive     Reply Quote
Admiral_Hawkes

Send message
Joined: 5 Aug 04
Posts: 106
Credit: 1,886
RAC: 0
Message 404 - Posted: 8 Aug 2004, 10:14:25 UTC - in response to Message 397.  
Last modified: 8 Aug 2004, 10:16:05 UTC

&gt; I have made a prelimary CPU-time/comparison page (based on 1000 trickles)
&gt; here:
&gt;
&gt; <A HREF="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/cpu.html">CPU Timings So
&gt; Far</A>
&gt;
&gt; I think I was a bit off in my original "spam beta invitation" email, a P4
&gt; 2.6GHz looks like it would complete in just under 3 weeks, not 4 weeks
&gt; (running "24/7")
&gt;
&gt;

AMD Athlon(tm) 64 Processor 3200+ Pentium 1 Microsoft Windows XP 2692.0 5265.0 767.48 976.56 2.042 18.385

I have this kind of CPU and I only have 2.3 s/TS. That's pretty much and I can't understand because I have no other background prozesses. SoCPDN gets as much RAM as it wants and 99% CPU time! What can i Do to raise my performance???

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>
ID: 404 · Report as offensive     Reply Quote
old_user412

Send message
Joined: 7 Aug 04
Posts: 18
Credit: 70,985
RAC: 0
Message 411 - Posted: 8 Aug 2004, 12:30:02 UTC

Getting 4.1 secs/ts with Athlon XP 2000+ Running Windows XP
ID: 411 · Report as offensive     Reply Quote
old_user1
Avatar

Send message
Joined: 5 Aug 04
Posts: 907
Credit: 299,864
RAC: 0
Message 416 - Posted: 8 Aug 2004, 13:18:42 UTC - in response to Message 404.  

&gt;AMD Athlon(tm) 64 Processor 3200+ Pentium 1 Microsoft Windows XP 2692.0 &gt;5265.0 767.48 976.56 2.042 18.385
&gt;
&gt;I have this kind of CPU and I only have 2.3 s/TS. That's pretty much and I &gt;can't understand because I have no other background prozesses.

That's exactly my CPU as well and I get 2.42 s/TS with 1GB of RAM. I think memory bandwidth is very important, unfortunately BOINC seems to give everyone the same #/measurement for memory speed, but I know my memory is slow (166?) -- I bought my AMD64 PC off of ebay! :-)

ID: 416 · Report as offensive     Reply Quote
Profile old_user419

Send message
Joined: 7 Aug 04
Posts: 7
Credit: 63,378
RAC: 0
Message 421 - Posted: 8 Aug 2004, 14:19:29 UTC

P4 3.06 GHz HT laptop,
512 MB RAM,
Windows XP Pro SP1
running 1 BOINC model + classic CPDN simultaneously =
3.88 s/TS
ID: 421 · Report as offensive     Reply Quote
old_user577

Send message
Joined: 8 Aug 04
Posts: 5
Credit: 154,634
RAC: 0
Message 422 - Posted: 8 Aug 2004, 14:29:20 UTC

P4 2.8 GHz HT, using one CPU
Linux 2.4.21smp4G (Suse 9.0)
2.35

Jürgen
http://www.wetterstation-porta.info
ID: 422 · Report as offensive     Reply Quote
jabba

Send message
Joined: 5 Aug 04
Posts: 4
Credit: 720,492
RAC: 0
Message 427 - Posted: 8 Aug 2004, 15:17:46 UTC

AMD T-bred @ 1804MHz, 256kB L2$, (XP2200+ ?)
768MB @ 133MHz SDRAM (KT133A chipset),
Linux 2.4.20
4.87s/TS

By the way. Running the ./viz takes one half CPU-time?! Is it really that heavy application? If so, one would be adviced to check the nice graphics only once and then... :)

Cheers,
Jani
ID: 427 · Report as offensive     Reply Quote
old_user245

Send message
Joined: 6 Aug 04
Posts: 3
Credit: 303,498
RAC: 0
Message 428 - Posted: 8 Aug 2004, 15:25:34 UTC

I get 1.86s/TS.

xp2500 mobile @ 2520MHz
10.5x240
Windows xp pro sp1
ID: 428 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,403,322
RAC: 5,085
Message 431 - Posted: 8 Aug 2004, 15:33:59 UTC - in response to Message 397.  

&gt; I have made a prelimary CPU-time/comparison page (based on 1000 trickles)
&gt; here:
&gt;
&gt; <A HREF="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/cpu.html">CPU Timings So
&gt; Far</a>
&gt;
&gt; I think I was a bit off in my original "spam beta invitation" email, a P4
&gt; 2.6GHz looks like it would complete in just under 3 weeks, not 4 weeks
&gt; (running "24/7")
&gt;
&gt;
Carl,

Cosmetic fix suggested. In your comparison web page, shouldn't the it be sec/TS instead of TS/sec since the lower numbers are leading the pack? Thanks.

George
ID: 431 · Report as offensive     Reply Quote
scm

Send message
Joined: 5 Aug 04
Posts: 9
Credit: 139,730
RAC: 0
Message 450 - Posted: 8 Aug 2004, 17:37:27 UTC - in response to Message 431.  

AMD Athlon 1700+ 384Mo Win XP
-&gt; 5.23 s/TS
ID: 450 · Report as offensive     Reply Quote
old_user1
Avatar

Send message
Joined: 5 Aug 04
Posts: 907
Credit: 299,864
RAC: 0
Message 454 - Posted: 8 Aug 2004, 18:07:03 UTC - in response to Message 431.  

&gt; Cosmetic fix suggested. In your comparison web page, shouldn't the it be
&gt; sec/TS instead of TS/sec since the lower numbers are leading the pack?

whoops, yeah, I just changed that. I also have it generate every hour until I get some more interesting pages up and work on the trickle/BOINC credit logic.


ID: 454 · Report as offensive     Reply Quote
Pconfig

Send message
Joined: 5 Aug 04
Posts: 84
Credit: 76,646
RAC: 0
Message 473 - Posted: 8 Aug 2004, 21:47:57 UTC

Hey carl it looks nice. You know, the PIV's have much better scores. I think that's strange cause i know that in the benchmarks the PIV score less then the Athlon series! That means that the intel pc's well get less credit for 1 wu!

It would be nice if you could include the computerid to find easily wich pc is yours!
ID: 473 · Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 . . . 6 · Next

Message boards : Number crunching : Client performance

©2024 climateprediction.net