I’m writing up some stuff on tiny numerical discrepancies in some hardware benchmarks I’ve been performing. At the moment I’ve chalked them up to pseudorandom numbers being introduced into the calculations by the algorithm in use or maybe slight differences in floating point arithmetic between the ostensibly identical compute nodes I’m testing.
I have written a little bit on why memory corruption is unlikely to be the culprit, and remembered something I read a while back about cosmic rays being a nontrivial source of – ahem – random bit-flips in RAM without error correction.
This gets me thinking – would horizontal vs. vertical orientation of RAM DIMMs change the likelihood of cosmic ray collision to any significant degree?
It’s my understanding that most cosmic rays / secondary shower particles that reach the ground come in at steep angles (less intervening atmosphere to potentially interact with), so I’m wondering if in the olden days anyone mounted RAM/other components edge on with respect to this preferred direction of attack…
(I recall reading Dan Rutter discussing the cargo cult computer configuration (CCCC!) that went on in days gone by, when people always span up hard disk drives on the same axis (with respect to gravity) they were formatted in. Very interesting, not sure how necessary it actually would have been)