But how in the world are you going to manage without ecc memory? You might have a bit get flipped by a cosmic ray once every 5 years of 24x7 computing.
There was an anecdote[1] about a Power Mac cluster at a university, which basically couldn't get past boot because the memory errors was so frequent.
[0] https://en.wikipedia.org/wiki/Cosmic_ray#Effect_on_electroni... [1] https://spectrum.ieee.org/computing/hardware/how-to-kill-a-s... [2] http://www.pld.ttu.ee/IAF0030/curtis.pdf
But even if true, so what? What are you doing on a workstation where a bit flipped every 3 days is going to cause a problem? Anything critical is going to have a software check. I just have never seen any legitimate argument for needing it in a workstation.
Not everyone crcs 100% of their structs.
If they don't, data is being slowly corrupted.
Hardware was at fault. Turns out suppliers push the truth a bit here and there about their chips, what is supposed to be an identical part # isn't always identical, parts from different manufacturers that are supposed to be interchangeable aren't, and technical manuals aren't always the best.
We had (what we thought was) rigorous memory testing in place at the factory, but under certain extreme conditions there was a 100% reproducible flip of a few bits in RAM. It was almost always the same bits, thankfully, that made it possible to track down!
Depends what bit flips. Given some of the file formats out there, the bit flip just before it is written to the disc might be fatal.
> Anything critical is going to have a software check.
I would love to see someone go through github and analyze if that is true.
I just cannot imagine why it would be acceptable to have bit errors when a preventative measure is available. If the industry would stop looking at it as a value add, the cost would come down and we all would benefit from a more stable platform.
What instability are you talking about? Has anyone experience any instability ever on a workstation that could have been attributed to a cosmic ray bit flip?
I wouldn't trade a 10% memory performance drag for ECC if the memory was cheaper than standard.
By definition, if the bits can flip in a way that wasn't intended then it is not stable. I'll take the 10% because I pretty sure getting to wrong 10% faster is not worth it to me.