UGM1, MCM1, CEP2... I'm confused as this is the UGM1 forum. Not aware of a boot and fail correlation on CEP2, just the large progress loss unless a boot is planned.

Klik did post a link to the Memtest86 app I've mentioned, though oddly the website name is memtest. It's good for both 32 and 64 bit platforms.

Surprise and incredulity is on the rise over the smallest thing not going the way we want it. My surprise was this message was never ever seen or reported before in the 10.7 years computing of WCG's existence, even more surprise the particular error was never documented on Google's internet, so really like to see the error log from the Result status page, if it contains anything. Certainly like to see the event log sequence from the stdoutdae.txt file, the before and after... an actual error code. Anything new that got written to the stderrdae.txt at the time of the incident. Both the files are in the BOINC data directory, path printed at top of Event log view after a client restart.

If a task fails during any phase without the ability to regress to a previous checkpoint, there's no recovery, no trusting the output, so yes it's the cheapest way to reprocess. If it fails the same way for a wingman, there's a reproducible problem, if not it was the rarest of flukes [IMO].

Beyond that, maybe a tech can look up what the message entails within the application code. It may be a UGM home made.