Error rates during double-check stage

[Greenbank]

Any news on the error rates that you've discovered so far?

One of my p4's has some memory problem. Prime95 torture test fails it quite early, sbfactor reports random low factors (i.e. < 10) on entries in SoB.dat, and a few other bizarre things.

However, from all testing I've done (comparing it to Xeon that works perfectly) the residues from various proth tests match perfectly.

Do you check to see if one client is responsible for lots of incorrect residues? Or at least see if one user (if they have multiple machines behind one IP address) is throwing out a large number of non-matching residues?

[EDIT]

Just ran through a few small n's again and no differences between the p4 and my pIII laptop. Must say that it's a pain in the a*** to get around that checksum value in the cache file (on linux), gotta love GDB though. :-)