 |
|
Message boards : Number crunching : What's your lowest s/TS?
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
Just as a matter of interest, how low s/TS numbers do you get on the very fastest Intel's and AMD's? I'm running an AMD64 3200 which is usually running at 2.27 s/TS. <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
 astroWXSend message Joined: 5 Aug 04 Posts: 1496 Credit: 95,522,203 RAC: 0 |
P4 HT 3.4, SuSE Linux 9.1 Best case: 2.4 & 2.5 sec/TS for simultaneous runs, yielding 1.23 sec/TS throughput. More typical for the box is closer to 1.3 sec/TS throughput. (Not overclocked.) "We have met the enemy and he is us." -- Pogo Greetings from coastal Washington state, the scenic US Pacific Northwest. |
|
geophi Send message Joined: 7 Aug 04 Posts: 2187 Credit: 64,822,615 RAC: 5,275 |
Without overclocking... AMD64 3400+ (2.4 GHz Winchester) In fastest run currently at 1.86 sec/TS AMD64 3200+ (2.0 GHz ClawHammer) About 2.15 P4 3.4 GHz (Northwood) Fastest model 1.36, Fastest two concurrent models using HT 2.30 and 2.49 AthlonXP 3000+ (166/333 FSB) 2.45 |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
@geophi It seems there are different AMD64 3200's. Mine runs at 2.2 GHz. Even stranger is that your's produces lower values :-) But then I don't really understand CPU's ... Can the s/TS vary depending on the model being run, or is it a (more or less) constant figure which only depends on the hardware and software in use? I'm currently running my first model, i.e. the first to not play up after 5 hrs. <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
Honza Send message Joined: 5 Aug 04 Posts: 390 Credit: 2,475,242 RAC: 0 |
> It seems there are different AMD64 3200's. Mine runs at 2.2 GHz. Even stranger > is that your's produces lower values :-) But then I don't really understand > CPU's ... Winchester vs. Clawhammer. Note the memory configuration (single vs. dual channel) and it's speed + latency makes a difference as well. > Can the s/TS vary depending on the model being run, or is it a (more or less) > constant figure which only depends on the hardware and software in use? I'm > currently running my first model, i.e. the first to not play up after 5 hrs. That's true; not eavery models are the same - i would say about |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
@Honza: You mean different models give roughly the same values? <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
Anybody using 3.6 - 4.0 GHz P4's? <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
geophi Send message Joined: 7 Aug 04 Posts: 2187 Credit: 64,822,615 RAC: 5,275 |
> @geophi > > It seems there are different AMD64 3200's. Mine runs at 2.2 GHz. Even stranger > is that your's produces lower values :-) But then I don't really understand > CPU's ... > Both my A64s are 754 pin with single channel memory controllers. The Clawhammer has 1 MB L2 cache instead of the 512 KB that the Winchester has, but that should make little difference to CPDN. More than likely the reason my 3200+ is faster than yours is memory. Mine has low-latency PC3200. Some Athlon64 systems are fitted with PC2700 memory. But even more likely is that the memory timings on your system are slower. Mine is running at 2-2-2-10 with 1T command rate, lots of motherboards default to 2T command rate and that can make a large difference. But other contributors could also be an exceptionally slow model, and other processes taking up CPU time. |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
> Both my A64s are 754 pin with single channel memory controllers. The > Clawhammer has 1 MB L2 cache instead of the 512 KB that the Winchester has, > but that should make little difference to CPDN. More than likely the reason > my 3200+ is faster than yours is memory. Mine has low-latency PC3200. Some > Athlon64 systems are fitted with PC2700 memory. But even more likely is that > the memory timings on your system are slower. Mine is running at 2-2-2-10 > with 1T command rate, lots of motherboards default to 2T command rate and that > can make a large difference. > > But other contributors could also be an exceptionally slow model, and other > processes taking up CPU time. Interesting. I do have PC3200 memory I believe, but that command rate business I have now idea about. Is that a BIOS setting? <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
Honza Send message Joined: 5 Aug 04 Posts: 390 Credit: 2,475,242 RAC: 0 |
> Anybody using 3.6 - 4.0 GHz P4's? I had a 3.6 and even 3.8GHz Presshot for testing. The faster one was giving about 20 days per 2 model with hyperthreading on, IARC. > You mean different models give roughly the same values? Yes, different models give roughly the same values. I remember UK_Nick posted such a table of model inter-variability. Having more complex models in near future than now in use Slab SM3 will take a bit longer to preccess. |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
> Having more complex models in near future than now in use Slab SM3 will take a > bit longer to preccess. Slab SM3 = Experiment 2, simulating 1950-2000? <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
Send message Joined: 26 Aug 04 Posts: 59 Credit: 438,133 RAC: 0 |
My AMD 3000+ is running at approximately 3.65 s/TS. It's a 333MHz FSB with 512KB L2 cache, and the ram is also running 333MHz (PC2700) - I believe they are 2.5 in timing (whatever it's called). Any idea why it's so slow compared to Geophi's 2.45s? |
|
old_user17525 Send message Joined: 13 Sep 04 Posts: 161 Credit: 284,548 RAC: 0 |
> > Slab SM3 = Experiment 2, simulating 1950-2000? > yes, see <a href="http://www.climateprediction.net/science/strategy.php#exp_2">here</a> for experiment strategy _________________________________ |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
> My AMD 3000+ is running at approximately 3.65 s/TS. It's a 333MHz FSB with > 512KB L2 cache, and the ram is also running 333MHz (PC2700) - I believe they > are 2.5 in timing (whatever it's called). Any idea why it's so slow compared > to Geophi's 2.45s? Well, couldn't it be because of the specs you just mentioned? His RAM and CPU are faster. (Not proportionally to the s/TS difference though, admittedly.) <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
Send message Joined: 5 Aug 04 Posts: 66 Credit: 2,146,056 RAC: 0 |
> > > > > Slab SM3 = Experiment 2, simulating 1950-2000? > > > yes, see <a> href="http://www.climateprediction.net/science/strategy.php#exp_2">here</a> > for experiment strategy > Except that Expt 2 uses a coupled oceam model, CM3, requiring up to an order of magnitude more processing. |
|
old_user17525 Send message Joined: 13 Sep 04 Posts: 161 Credit: 284,548 RAC: 0 |
oops, sorry. (next time I'll just do one thing at a time!) _________________________________ |
|
Send message Joined: 16 Oct 04 Posts: 692 Credit: 277,679 RAC: 0 |
Ediited out. Sorry missed that someone else had answered. Visit BOINC WIKI for help And join BOINC Synergy for all the news in one place. |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
> Except that Expt 2 uses a coupled oceam model, CM3, requiring up to an order > of magnitude more processing. So, when Experiment 1 is done, will we be doing SM3 or CM3? <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
|
Send message Joined: 16 Oct 04 Posts: 692 Credit: 277,679 RAC: 0 |
Experiment 1 is SM3 (slab model). Both exp1 and SM3 apply to both current work and the next step which is adding a sulphur cycle. Experiment 2 will use a fully dynamic ocean ie CM3 (coupled model). I expect that experiment 1 and 2 will overlap to a certain extent. Visit BOINC WIKI for help And join BOINC Synergy for all the news in one place. |
|
Send message Joined: 5 Feb 05 Posts: 30 Credit: 77,091 RAC: 0 |
> Experiment 1 is SM3 (slab model). Both exp1 and SM3 apply > to both current work and the next step which is adding a > sulphur cycle. OK, thanks. Just was a little confused by <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=1832#8800">Honza's reply</a> Anyone know of any risks of changing the memory command rate from 2T to 1T? <br><br>_______________________<br><a href="http://doodle.naylor.se/">Doodle Theme for Firefox</a><br><a href="http://david.naylor.se">My Web Gallery</a><br><a href="//naylog.blogspot.com/">My Blog</a> |
©2024 cpdn.org
