Titan V may generate false results.

[MisterBitcoin]

Different results for the same test (proteine and enzyme simulation), well thats not good.

Source.

[VictordeHolland]

I wouldn't be surprised by this IF it was a consumer/gaming card, not a 2000-3000 €/$ card....

[retina]

Non-ECC memory so such things should be expected. And it isn't just this card that will exhibit the problem, all systems without ECC memory are affected.

[ixfd64]

Probably a dumb question, but does ECC memory reduce the error rate to near zero, or just by a portion?

[Madpoo]

Originally Posted by MisterBitcoin

Different results for the same test (proteine and enzyme simulation), well thats not good.

Source.

From the article:

After repeated tests on four of the top-of-the-line GPUs, he found two gave numerical errors about 10 per cent of the time.

So essentially an error rate of 5% total for the 4 machines, which, hey, is about what we see here at GIMPS.

[Madpoo]

Originally Posted by ixfd64

Probably a dumb question, but does ECC memory reduce the error rate to near zero, or just by a portion?

I'll go out on a limb and say it reduces it to near zero, at least for Prime95.

In all of my tests on ECC enabled systems, never had a bad result (and I only did double-checks, so... there you go).

There are also other users out there who use ECC systems almost exclusively (on AWS instances, or whatever else) and those also have an impeccable history.

Unfortunately Primenet isn't tracking whether or not ECC is in use on the systems (hey, feature request!) so this is anecdotal evidence.

[retina]

Originally Posted by ixfd64

Probably a dumb question, but does ECC memory reduce the error rate to near zero, or just by a portion?

The reduction ratio would be approximately this:

Going from one single error in the gigabytes of data killing the entire run, to needing three (or more) concurrent errors all clustered into a single 72-bit (64 bits of data + 8 bits of checksum) memory address.

[ixfd64]

I see. So that would make the chance of an error very, very low, then.

[kriesel]

Originally Posted by ixfd64

Probably a dumb question, but does ECC memory reduce the error rate to near zero, or just by a portion?

This reference seems to indicate memory dominates because it's most of the area of the active system devices, but transmission lines and logic circuits are also affected. Cosmic rays dominate. https://en.wikipedia.org/wiki/Soft_e...oft_error_rate

[retina]

Originally Posted by kriesel

Cosmic rays dominate.

Do you have a reference for this? I think that is not actually correct. Most cosmic rays are stopped well before reaching ground level. I think most of the soft errors are actually caused by radioactive impurities in the plastic packaging.

[airsquirrels]

Originally Posted by retina

Do you have a reference for this? I think that is not actually correct. Most cosmic rays are stopped well before reaching ground level. I think most of the soft errors are actually caused by radioactive impurities in the plastic packaging.

The Wikipedia arrival disagrees, but I don’t know if that counts as a reference.

There’s a solution though. “Burying a system in a cave reduces the rate of cosmic-ray induced soft errors to a negligible level. ”

Perhaps we should become GIMPs miners and move to the underground since were losing Madpoos double check powers.