Originally Posted by kweechy
Ehhh I think ECC is like an old wives tale these days. The smaller the transistors get, the less likelihood for cosmic ray bit flips as well.
I'll just build a Faraday Cage to put my office in!
Seriously though I've never personally encountered any non-ECC problems yet in this lifetime so far. I've heard of one story from the CTO at our studio, but that's it.
If I were running a server, I'd use ECC to protect the files and stuff. But if the computers are just worker nodes, they don't need it. Worst case scenario, you'd lose one work unit to cosmic rays, but nothing terribly detrimental.
And again, never seen it happen yet personally.
Cosmic rays are not the only cause of bit flips though.
The smaller and faster transistors get.... the greater the chance of leakage/crosstalk/noise/bit flips assuming all else equal.
In addition, you may not notice or detect a bit flip. Not all bit flips cause a memory error or system failure.
What's the difference between 62523261554 and 45343392370? One bit.
What's the difference between $62,523,261,554 and $45,343,392,370? A lot of money.
If you're running financial models and use the output for investment decisions, the risk of not having ECC is not worth it.
You mention "studio" as in art/graphics studio? A bit flip may not be a big deal if a single pixel is wrong.
Originally Posted by BizzareRide
Expensive RAM is far over rated. There's an article that showed that cheaper RAM didn't give anything up to the more expensive variety. RAM is fast enough as it is. The real bottleneck in the memory hierarchy is the main storage. Even with the advent of SSDs, there's still a huge chasm between RAM and solid state memory, though probably not as large as the one between RAM and optics.
This is actually one of the major reasons Intel is invested in SSDs... to develop faster storage systems that won't bottleneck CPUs.