In the FAQ, item 6.2 says:

The Disk and Work tabs do a great deal of communications with the project code. Neither of these should be your normal display tab, even when minimized. A 2 GHz computer can lose 1% of processing keeping these tabs updated. The overhead is much greater for slower computers (a P166 can lose 50% of processing).

(my client doesn't have a Work tab - this was the name for the Tasks tab in version 4.xx or 5.2.x?)

My question is, does the advice still apply, given that it was (apparently) written some time ago?

I am now leaving my client open on the Projects tab, rather than the Tasks tab.

Way back then, computers were a lot slower, and any little improvement to let most of the cpu power go to modelling, rather than house keeping, was taken.

These days, the cpu architecture is a lot different, and the house keeping is a smaller percentage. The main thing that takes up cpu time is letting any graphics display run all of the time, or using the cpdn screen saver, instead of a generic saver. I used to use "blank" for the saver.

However, you do have two other problems:
1) Not really enough memory for running climate models - 2 Gigs would be better.
2) Your computer is crashing everything with the error:
couldn't start app: CreateProcess() failed

This could be a permissions problem.

Thanks for this, although given the vintage of my machine, the notes in the FAQ may still apply! I'm pretty sure my chip needs all the cycles it can get...

Thanks also for pointing out the other issues, I can deal with a long-running task, but not if it's crashing. My machine did crash on previous tasks, with the error you mention, but the current task shows status "in progress" and an STDERR of "--". The tasks tab also shows 2% complete (slowly increasing). It seems to be running OK... can you see an error on the "in progress" task?

Memory, that's not good news. The machine is used only for BOINC (it's not my main computer, I just leave it running in the corner). I was hoping I could get away with a low-spec machine due to this. Are you saying that's just not going to happen? It seems to be handling it so far - but maybe it will run out of memory as the task progresses? The machine does meet the minimum spec listed here. But maybe that page is a little optimistic?

All advice welcome, apologies for my newbie questions. :-)

The machine does meet the minimum spec listed here.

However, the memory listed on the page for your computer shows only 702Mb, presumably the rest going to your graphics. I found that going up to 2GB/core greatly reduced the number of crashes on my machine though as I haven't used windows this century I don't know how much your OS gobbles up out of the 702MB that is left by the graphics. I don't know if there is a way of cutting down how much the graphics uses by 100MB as I suspect that may make a difference.

Dave, thanks for the suggestions. Next reboot, I'll see if I can tweak the BIOS to recover some RAM from the graphics card. But that's not going to take me to 2Gb!

As it happened, my task died with a runtime error. There were two dialog boxes, a C++ error from BOINC - and a message from Windows saying �Your system is low on virtual memory, windows will now increase the size of your paging file....". So I guess Les was right. :)

Never mind, it was a good learning experience. Nice to meet you all as well. I do have better hardware available. I do like the low-power, low-noise and small size aspects of the machine I'm using (an old laptop). But if it can't handle the job, I'll need to reconsider.

I have been running these climate WU on a 1GB (with ample virtual memory) formally Windows XP now Lubuntu desktop for many years and they complete. Most of the time I run them concurrently with WCG. Like many others my WU result list is littered with download errors and incompletes (short models, PNW). When running this project on Windows I do set my virus scan to exclude the BOINC data directories and I pause the calculations when a full disk virus scan is scheduled to run. I am also careful to avoid shutdowns near the 25,50,75 percent complete points. Of course it can take weeks or months to complete the long models but it is a 10 year old "value" class machine.