Memory requirement for CEP is about 1 GB PER wu, suspect you are running into a low memory problem

Hi Aperture_Science_Innovators,
Speaking of memory, how large is your virtual memory? Task Manager just shows the working set, which is usually a tiny portion of the total virtual memory consumed.

Lawrence

Hi Aperture_Science_Innovators,
Another good idea shot down! 12 GB virtual memory is about right for the load that Task Manager showed. I suppose that you have Windows set to automatically increase the VM Cache size if required. Which is how you should have it set.

Right now I am out of suggestions.

Lawrence

I think you have enough physical memory to run more than 12 CEP2 jobs, though if you're running WCG projects on all 16 cores there won't be much to spare. But I think your problem is something other than amount of memory:

1. When a CEP2 WU starts, it immediately unzips several thousand (!) small data files (refrains from comments re FORTRAN programmers and their mastery of sound computer science principles). This causes a lot of overhead for the O/S, as well as the storage device hardware. If a number of CEP2 WUs try to start up simultaneously, the system can freeze for many seconds, possibly because file creation is probably a single-threaded process, and creating all those files just overwhelms it. In that situation, other BOINC tasks can experience timeouts, causing them to exit. "No heartbeat from client" is a common error message.
I haven't looked at how much O/S overhead there is in accessing all of these files once the main part of the CEP2 calculations are underway, but with so many files it must be considerable.

2. Page faults: I suggest you have a look at this parameter in your task manager (View >> Select columns). CEP2 produces heaps of these, and I suspect that they too are single-threaded in the O/S.

Hope that helps.

BTW, the CEP2 science team are working on upgrades to the program. No details are known, but fingers crossed they may be addressing some of these problems.

Refer to what Rickjb and Lawrence already wrote... yes, the IO to a 5400 RPM spinning drive will add subtract from the success. Staggered starting of multiple CEP2 tasks is recommended, unfortunately hard to automate [nothing in BOINC to control this], but once they're all running past the first checkpoint, chance of 2 or more finish/start simultaneous will be minimal. My 8 core has 8GB DDR3 and becomes inefficient, particularly when used, but it 'can' do 8 in hands off mode.