I ran CPDN for a while, until I went into an \"iceage\", which I was told to abort as the WU would never finish.

Recently, I thought I would give it another go. But, when I attached I got what seems to have been a 387 hour WU. So, I said no thanks.

I just tried again and got what seems to be over 2900 hours WU. Is that accurate? Am I misreading something?

>>RSM

---

>>RSM

The climate work units have always been long cycle work units. To offset that, the credit trickles are handled on a daily basis. It used to be that 2000+ hour work units were standard. Now there are different types of work units depending on choices you configure for application preferences. There are three options there. For me, I\'ve shifted to work units that run in the 250 hour range (the HADSM3 application), the HADCM3 work units would run in the 700 hour range for my processors (mostly AM2 4200 to AM2 6000). The HADAM3 applications would run about 150 hours -- but this one requires the workstation to have 1.5G of RAM or more and I have some with only 1G so I don\'t use that.

But there shouldn\'t be 2000 hour long applications from Client these days -- what sort of CPU are you running?

I ran CPDN for a while, until I went into an \"iceage\", which I was told to abort as the WU would never finish.

Recently, I thought I would give it another go. But, when I attached I got what seems to have been a 387 hour WU. So, I said no thanks.

I just tried again and got what seems to be over 2900 hours WU. Is that accurate? Am I misreading something?

>>RSM

---

On a moderately fast computer the HADAMs usually take about 15 days, the HADSMs about 3 weeks and the HADCM 160-yrs rather more than 3 months running 24/7. The HADCM 80-yr models take about 7 weeks but you can\'t select 80 or 160 years; you get what is next in the queue.

You can select the type of model you want next in the CPDN preferences section of your account. The HADAMs are proving to be the most reliable; hardly any develop defects.

Yes, they\'re all massive tasks compared with other projects. But, as Barry says, you do get your credits as you go. And it doesn\'t matter if you take much longer to complete them as the CPDN server ignores missed deadlines.

The CPDN READMEs linked in my signature give lot of hints on how to keep models going to completion. In the collection about crashes and problems I particularly recommend item #6 by Mike.

The crunched models, as well as being used by the CPDN researchers, are being offered by Milo to climate scientists world-wide. At least half-a-dozen groups have already downloaded sets of our results. So I\'d say the crunching is well worth while in spite of the initial shock when you see what the jobs consist of.

Another good thing is that because each task within a workunit can be expected to turn out slightly differently due to differences between machines and the Lorenz \'butterfly\' effect, they will all be used. So unless you get a defective model there\'s no wasted crunching. In the case of HADCMs I believe that even if a model develops a defect, its data up to that point can be used.

---

Cpdn news

... over 2900 hours WU

That\'s nothing compared to what the physicsts would really like.
At their \"ideal\" resolution, using the fastest desktop computer currently available, one of these ideal models would take about 2,500 REAL years to complete.

So what is being run at present is very much a compromise between \"wanted\" and \"realistic\".

Several posts about trickles that were posted at this point have been moved to this thread

Well, if they want to wait 2500 real years, they don\'t need to run a model, just collect the actual event driven data. Seems a rather foolish \'really like\' scenario for predictive software models not to complete until after the fact.

... over 2900 hours WU

That\'s nothing compared to what the physicsts would really like.
At their \"ideal\" resolution, using the fastest desktop computer currently available, one of these ideal models would take about 2,500 REAL years to complete.

So what is being run at present is very much a compromise between \"wanted\" and \"realistic\".

[/quote]

---

2,500 years is what the models would take on desktops.
They actually want the models in a time frame of a couple of months, the same as the current models.

The really hi-res program that Carl was working on just before he left would/will help a bit in this direction, but it needs 4.5 Gigs of ram.
And we\'ve lost our insider info about that project.

Hmm, do the models take into account the amount of heat produced to run the models? <smile>.

2,500 years is what the models would take on desktops.
They actually want the models in a time frame of a couple of months, the same as the current models.

The really hi-res program that Carl was working on just before he left would/will help a bit in this direction, but it needs 4.5 Gigs of ram.
And we\'ve lost our insider info about that project.

---