I've been watching the graphic globe occasionally on the BOINC version of the model. About a week ago I watched closely how long it would take to do it's work, and it seemed to me that it was simply slower than how the older model software would run.

After doing a 60 second test with both I found that a model under BOINC runs 150% slower than it does under the older software. I still have both installed on my computer, but only BOINC is installed as a service and normally only BOINC is running now.

Because of this significant difference I am considering going back to the old software. However the last time I checked automatic model uploads weren't working with the older software, and also auto downloads of new models wasn't working either.

In sixty seconds under BOINC I get 16 timesteps, whereas under the older software I get 24.

Are there steps under way to correct this great descrepancy with BOINC, or steps to allow people who have AMD processor computers to be able to do auto downloads & uploads using the older software?

Using the slower BOINC seems kind of a waste of time for me & for the project.

Here's the benchmark under boinc:

1903 double precision MIPS
4588 integer MIPS

Assuming that the older model running software is even still usable, what would be the preference of the project organizers in cases like this - to drop a BOINC running model mid stream & go back to the old immediately, or to wait until 1/3 of a run is complete (I'm on the first of the three on the AMD XP CPU computer) before going back?

---

From what I have seen and read, if you have the graphical globe displaying under boinc, especially with the star field showing the processing of the model is significantly reduced. This is as the graphic is actually quite CPU intensive and will give similar results to what you found.

Others have found that the boinc version actually runs faster, especially on Intel processors.

Another thing: I think that the latest CPDN models are actually including more atmospheric / oceanic chemistry /phsyics /interactions so they may be slower. Or this may be a plan in the works, I haven't been able to access the other forums to check lately. All new aspects to the simulation I believe will go into the boinc version, not the Classic version.

Third, each individual parameter set runs at a different rate and if you check your trickle times for different runs with the same model (diff parameters) you will notice that there can be quite a difference.

Anyway, hope that this helps you decide where to allot your spare processing cycles.
DaveN

---

150% is a large amount and i don't think this is the case in general.
If you are running the visualization particularly the cloud scene in full screen mode, then yes this does take up a significant amount of cpu time.
I think Carl added an option to strip the could levels because of this.
( key 1,2 ...)

the classic version uses a COIN GL toolkit that performs low level rendering
whilst the BOINC version uses vanilla OpenGL calls.
If you'd like to give as much cpu time to the worker while the graphic window is active and not have to revert to the classic client, you can switch the default scene to the wireframe. key' 7'

---

While the BOINC client may be a little slower on your CPU than classic, it should be on the order of 0 to 5% slower, not 50%. Tolu is right, it is likely the visualization which is slowing you down that much in BOINC. Run it through a complete trickle while not using the graphics, then compare that trickle duration to the duration untder classic. It should be nowhere near 50% longer.