Hello, for along time ive ran Boinc with the World Community Grid and it had no problem using all 8 cores and lots of RAM.

Ive decided to move over to this after reading about the good work the hadcm3 model was doing.

I allowed all the World Community Grid computations to finish, however the Climate Prediction project is only ever running one instance using 12-13% of the CPU.

Everything is set to allow Climate Prediction to use 100% of the CPU 100% of the time and allowance of 80% of the 18Gb of RAM which i have.

The CPU is an i7 930 overclocked to 3.8Ghz and it used to chew through World Community Grid things very quickly.

Can anybody advise how to get another 7 instances of the hadcm3 engine to run, or allow the one instance to use all cores.

Thanks,
Stuart

1) There's very little work at the moment.
It's released in batches, and then there's a long wait for data to be returned and analysed before more is released. At present, what little does show up is as a result of another computer failing to successfully complete a model, which then gets resent.

2) There are NO multicore models here.

3) Overclocking is not a good idea here.
Climate models consist of weather over a very long period. Weather is what is known as a chaotic system, which can change quickly and randomly.
Overclocking decreases the time needed for a processor to access the FPU, which is heavily used by these models. As a result, the results may well be something other than what they should.

---

Backups: Here

Just checking Les is correct in his assumption that it is because your computer has only has one task at the moment? If it has more than one task and only one is running at a time, the problem is something else.

Hi, i see - ill enable the World Community grid stuff to pick up the slack resources. Room is cool without the CPU running at 100%

I dont think any of the boinc things run multicore, they just run multiple instances together. Even Adobe render things work the same way.

I had hoped to see 8 instances of the model running but perhaps not if that much work isnt available.

On the subject of the overclock, i would have assumed the calculation will be irrespective of CPU speed. If i do 1+1=2 at any speed ill always get the same answer. If there is an error due to the overclock, the system will lock. its a very stable overclock.

Also, the CPU clock speed would naturally fluctuate with energy savings where it downclocks at low load and it naturally overclocks a few cores with the Intel TurboBoost when the core loading is uneaven.

With that all in mind, id have assumed there should have been no effects on the instruction being executed.

Dave, i was just running the one task. Now back fully loaded with tasks from other projects.

1 + 1 does equal 2.

But these calcs are for things such as sulphur pollutants, parts per million or less, and water vapour, a large % of the total. And lots of others.
The program, close to a million lines of floating point FORTRAN, needs time to get the correct data from the FPU.

Not to worry; if your models are statistically suspect, the researchers won't use them. Just like choosing fruit and veg at a shop: pick what looks good, and ignore what doesn't.

Been there, discussed this, many times.

******************

from your original post:

or allow the one instance to use all cores.

At least 2 projects used to run multi-core, but Acqua@home has ended.

---

Backups: Here

1 + 1 does equal 2.

But these calcs are for things such as sulphur pollutants, parts per million or less, and water vapour, a large % of the total. And lots of others.
The program, close to a million lines of floating point FORTRAN, needs time to get the correct data from the FPU.

Not to worry; if your models are statistically suspect, the researchers won't use them. Just like choosing fruit and veg at a shop: pick what looks good, and ignore what doesn't.

Been there, discussed this, many times.

******************

from your original post:

or allow the one instance to use all cores.

At least 2 projects used to run multi-core, but Acqua@home has ended.

Hi,

Thanks interesting, going to have a read up - id never considered over clocking as likely to affect a calculating.

I do alot of 3d graphics rendering, particle/fluid simulation in the likes of Adobe After effects, where effects packages run into Gb of programming, each frame may be 250MB in size. These things predict movement of billions of elements across thousands of frames so it may be similar in some respect to these models.

With this, i know, with the same settings applied, i will get the same result over and over again irrespective of the PC i run it on.

Anyway, i dont want to drag up a past subject, as i say - ive never considered it so i cant discuss it with any info with respect to boinc etc.

Thanks,
Stuart