climateprediction.net home page
Whats a good average sec/TS and other things...

Whats a good average sec/TS and other things...

Message boards : Number crunching : Whats a good average sec/TS and other things...
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile old_user206071
Avatar

Send message
Joined: 29 Oct 06
Posts: 14
Credit: 99,628
RAC: 0
Message 27832 - Posted: 12 Apr 2007, 16:17:52 UTC
Last modified: 12 Apr 2007, 16:18:30 UTC

I just downloaded a coupled model. [Its my second actually, my first model had an error after 160 hours or so...] Havent reported my first trickle (whenever this might be...) Its at 0.08% after 1hour 50min...

Anyway, whats a good average sec/TS? I have an AMD X2 4600+ EE @ 2.4GHz...

I\'m about to finish a SAP with 22.4 sec/TS, but I guess thats different. :?

It has 4147560 Timesteps, how many TS are one Trickle actually? And how many credits per trickle are granted? (No rounding please!)

Thanx in advance!

(And yes, I searched this message board ;)


My NEW BOINC-Site

Why people joined BOINC Synergy...
ID: 27832 · Report as offensive     Reply Quote
Profile Strathpeffer
Avatar

Send message
Joined: 9 Jan 07
Posts: 497
Credit: 342,899
RAC: 0
Message 27836 - Posted: 12 Apr 2007, 17:18:20 UTC

Hi, you may find this thread useful ;-)
Visit the Scotland team
ID: 27836 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 27837 - Posted: 12 Apr 2007, 17:19:40 UTC


My overclocked AMD X2 gets 1.8s/ts or thereabouts, but most X2s get around 2.0-2.3s/ts depending on their frequency and their memory speed.

Core2Duos will get around 1.1 - 1.3s/ts, P4s get around 2.2-2.5 s/ts.

One trickle is one model year (unlike SAP, where one trickle is 6 days), hence you get 160 trickles in the lifespan of the model.

There are 25920 timesteps per trickle. I get two trickles per model per day on my AMD.

0.01 credit per timestep (hence 259.2 credits per trickle). SAP gives 0.1 credit per timestep, but since their timesteps are about 10 times slower, that\'s fair enough.
I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 27837 · Report as offensive     Reply Quote
Richard Haselgrove

Send message
Joined: 1 Jan 07
Posts: 925
Credit: 34,100,818
RAC: 11,270
Message 27838 - Posted: 12 Apr 2007, 17:20:42 UTC - in response to Message 27832.  

I just downloaded a coupled model. [Its my second actually, my first model had an error after 160 hours or so...] Havent reported my first trickle (whenever this might be...) Its at 0.08% after 1hour 50min...

Anyway, whats a good average sec/TS? I have an AMD X2 4600+ EE @ 2.4GHz...

I\'m about to finish a SAP with 22.4 sec/TS, but I guess thats different. :?

It has 4147560 Timesteps, how many TS are one Trickle actually? And how many credits per trickle are granted? (No rounding please!)

Thanx in advance!

(And yes, I searched this message board ;)

Some specific figures to start you off:

There are 25,920 timesteps in a trickle, and you get 259.20 credits for that trickle - so exactly 100 timesteps = 1 credit.

I thought there were 160 trickles in a full run, making a total of 4,147,200 - some rounding errors must have crept in after all.

[Climate modellers will express the above figures as 3TS per model hour, or 72 per model day, multiplied by a \'notional\' 360 day model year: one trickle is also one model year]

As to speed - I\'m getting between 2.4 and 2.5 sec/TS on an Intel Xeon @ 1.86GHz: that means approaching 4 months CPU time to complete each model. Up to you to decide whether that\'s a \"good average\".
ID: 27838 · Report as offensive     Reply Quote
Profile old_user206071
Avatar

Send message
Joined: 29 Oct 06
Posts: 14
Credit: 99,628
RAC: 0
Message 27840 - Posted: 12 Apr 2007, 17:35:28 UTC

Thanx @ all!

With this I can update my excel sheet, so after 2:45 hours and 0.118%, there are still 0,507% (or appr. 10hours) left to my first 259.2 credits :)

Its sec/TS is currently 2.02, I hope it\'ll speed up a bit as soon as my SAP is finished. I read that the sec/TS is slower if two climate models run together... (No idea if its the same with a CPDN running with a SAP, though...)

Thanx again! :)


My NEW BOINC-Site

Why people joined BOINC Synergy...
ID: 27840 · Report as offensive     Reply Quote
Profile old_user206071
Avatar

Send message
Joined: 29 Oct 06
Posts: 14
Credit: 99,628
RAC: 0
Message 27842 - Posted: 12 Apr 2007, 17:54:20 UTC

Silly me, as mentioned above this is my second CPDN. The first actually failed after 300 hours (not 160), but its average sec/TS was 1.90 (not sure if it was overclocked at that time, though). Hmmm....


My NEW BOINC-Site

Why people joined BOINC Synergy...
ID: 27842 · Report as offensive     Reply Quote
Profile astroWX
Volunteer moderator

Send message
Joined: 5 Aug 04
Posts: 1496
Credit: 95,522,203
RAC: 0
Message 27844 - Posted: 12 Apr 2007, 18:06:58 UTC

The slight slowing was due to contention for the Hard Disc, I think. Version 5.40 has Disc I/O reduced by more than 90%, so it should be less of an issue.

The on-die memory controller is a benefit for AMD CPUs. (It\'s on the motherboard for Intel CPUs.)

For what it\'s worth, 5.15 Runs on my three A64 X2 4400+ boxes range from 1.88 to 1.98 s/TS.
The 5.40 Runs range from 1.86 to 1.89 s/TS, with one oddball at 1.95. (Given that all run in pairs, the slowpoke must be encumbered by a heavy parameter set.)

Two of the machines have Kingston Hyper-X and one has Patriot RAM. None overclocked, but all are set to best factory spec and 1T timing, Dual Channel. (2-3-2-6 1T for the Kingston, 2-3-2-5 1T for the Patriot.) I\'m given to understand that good memory matters a lot with an AMD X2; it should really make a difference with such long Runs as we have here.

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.
ID: 27844 · Report as offensive     Reply Quote
Profile old_user206071
Avatar

Send message
Joined: 29 Oct 06
Posts: 14
Credit: 99,628
RAC: 0
Message 27908 - Posted: 14 Apr 2007, 20:57:23 UTC

After having finished the SAP wu successfully, CPDN speeded up. Its now at 1.87 sec/TS... (X2 4600+ at 2.46GHz, 2x 1GB DDR2-667 DualChannel CL5)... :)


My NEW BOINC-Site

Why people joined BOINC Synergy...
ID: 27908 · Report as offensive     Reply Quote
Profile old_user206071
Avatar

Send message
Joined: 29 Oct 06
Posts: 14
Credit: 99,628
RAC: 0
Message 28234 - Posted: 28 Apr 2007, 8:38:14 UTC

AMD X2 4800+ @ 2.85GHz: 1.60sec/TS


My NEW BOINC-Site

Why people joined BOINC Synergy...
ID: 28234 · Report as offensive     Reply Quote
Profile Pooh Bear 27
Avatar

Send message
Joined: 5 Feb 05
Posts: 465
Credit: 1,914,189
RAC: 0
Message 28244 - Posted: 28 Apr 2007, 12:28:10 UTC - in response to Message 28234.  
Last modified: 28 Apr 2007, 12:28:41 UTC

AMD X2 4800+ @ 2.85GHz: 1.60sec/TS

My Core 2 Duo 2.4G (stock) does 1.60.

ID: 28244 · Report as offensive     Reply Quote
Steinar1965

Send message
Joined: 4 Sep 06
Posts: 79
Credit: 5,583,517
RAC: 0
Message 28264 - Posted: 29 Apr 2007, 8:29:46 UTC - in response to Message 28234.  

AMD X2 4800+ @ 2.85GHz: 1.60sec/TS

But that is with only one model running then? Guess you will see a small drop in speed when running two. Have you clocked your machine?
ID: 28264 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 28270 - Posted: 29 Apr 2007, 11:08:57 UTC

My old socket A AMD 1.33GHz whose CPU is probably fractured has been underclocked by 25%. Very good solution as it prevents the electronics part of the computer crashing the moment the climate model starts. It\'s now doing 8sec/TS and crunching steadily on a model that\'s a year old. The warm weather is also good for it as this expands the CPU and helps it hold together in one piece.

The Intel dual core 3600 with 800MHz RAM has 2 models both doing 1.56sec/TS.


Cpdn news
ID: 28270 · Report as offensive     Reply Quote

Message boards : Number crunching : Whats a good average sec/TS and other things...

©2024 climateprediction.net