A couple of weeks back I ran into a similar issue that might be related. Whenever I used network rendering to do batch rendering of a row of mxs (single machine, not network), I noticed the following frustrating behavior:
I would set up the render cue in monitor, start the rendering and then leave the machine. Then, when I came back a couple of hours later, I often would find that many mxs that hadn’t yet started rendering when I left, had only rendered to a ludicrously low SL (usually below 5), while having each maxed out their allotted render time. I could only think of one possible reason: Some other process with a higher priority than Maxwell seemed to have maxed out all cores while I was away.
My first thought went to malware, maybe some bitcoin miner or something similar, that would wait until the screen saver went on before becoming active. So I disabled monitor-turnoff, opened the task manager with the ‘Details’ page open. After I waited for about 10 minutes, a new process suddenly appeared in the list: System, rundll32.exe with the command line ‘C:\Windows\system32\rundll32.exe invagent.dll,RunUpdate’. This process would max out one core while maxwell.exe went down to 0% CPU load.
Turns out this is a gift from Microsoft that found its way on my machine via a security update. What I can gather from the KB entry (KB2976978,
https://support.microsoft.com/en-us/kb/ ... wsignin1.0), they are ‘gathering telemetry’ in preparation of the Windows 10 launch. Thanks for that.
After I uninstalled and deactivated this specific update, the issue went away.
Still, why would maxwell.exe be reduced to near zero CPU usage by another process that maxes out only one core? That should not be happening.