World Community Grid - View Thread

I haven’t seen it with WCG but certainly with Rosetta, where the memory usage is variable within a task over time, the tasks start with a low requirement and then, when the memory load increases, the machine runs out of both physical and swap then crashes.

But than either increasing the swap space and/or lowering the allowed RAM usage for BOINC is the correct solution (and of course disabling ARP), anything else will stop working as soon as the memory usage of the tasks increases. And than of course this might be the case with Rosetta, but here I have 14 MCM tasks running with progress somewhere between 0 and 100% and all of them use 38-40MB RAM. Total RAM usage is 7,3GB, 1,8GB of it for one Einstein's O3 alone. So 16 MCM in on a system with 8GB should work without any issues, there will be some swapping, but the entire system definitely should not crash.
Like I said, if the system is crushing within few minutes, simply open the task manager and watch the memory usage. No point in doing changes before you know what's the issue. In the same time you can open HWiNFO and watch the temperatures and clocks.

The fact that reducing the number of concurrent tasks has stopped the crash suggests that something like this may be happening here.

Reducing the number of concurrent tasks will also eliminate overheating, slightly unstable overclocks and/or undervolting or issues with failing hardware simply because the system isn't pushed that hard resulting in lower power draw and with that slightly higher or at least more stable CPU voltage.