X

Microsoft Research: laptops more reliable than desktops, faster CPUs fail quicker

For reasons known only to the great churning of the internet, a paper published in April 2011 and presented at EuroSys 2011 has only just bubbled to the top of web consciousness and looks into consumer hardware failures across a million PCs.

Craig Simms Special to CNET News
Craig was sucked into the endless vortex of tech at an early age, only to be spat back out babbling things like "phase-locked-loop crystal oscillators!". Mostly this receives a pat on the head from the listener, followed closely by a question about what laptop they should buy.
Craig Simms
2 min read

For reasons known only to the great churning of the internet, a paper published in April 2011 and presented at EuroSys 2011 has only just bubbled to the top of web consciousness and looks into consumer hardware failures across a million PCs.

(Credit: Seagate/CBSi)

The study was limited to CPU, RAM and disk failures that caused software crashes on Windows via the in-built WER tool. While some results (like disks becoming less reliable with the age of the machine) were obvious, quite a few interesting results came about:

  • Laptops are more reliable than desktops — they are between 25 and 60 per cent less likely to crash. This doesn't take into account usage

  • Faster CPUs become faulty more quickly than slower CPUs

  • A slightly overclocked CPU will "significantly degrade machine reliability" (although it's unclear if Turbo Boost-style auto overclocking was covered)

  • Slight CPU underclocking improves reliability dramatically over running at rated speed, at 39 to 80 per cent less likely to crash

  • CPU and RAM in pre-built machines (Dell, HP et al) have lower failure rates than custom built machines. Disks stay the same

  • While both AMD and Intel CPUs were tested for reliability, and one outperformed the other, the study annoyingly doesn't say which

  • Younger systems are more likely to have CPU failures, but there's no evidence shown of burnout

  • Adding more RAM reduces the risk of disk failure, although increases the rate of CPU failure, and minimally increases the chance of RAM failure

  • Age of system has no effect on RAM failures

  • Up to 1 in 190 machines crash over an 8 month period due to hardware failure

  • After the first crash, likelihood of the same crash re-occurring increases by up to two orders of magnitude

  • Recurrent failure will happens in 97 per cent of machines within 10 days of the first failure

  • 15 to 39 per cent of machines exhibited intermittent faults.

The recommendation from the team involved? Build a more fault tolerant operating system.

For the full stats, we'd recommend reading the paper on Microsoft Research's page.