[3geofftech]

Hi,
Looking for similar experiences or comment here - we supportmany power 5 and power 4 servers. No problems with the p4s but we are finding that with the p5s :
a) The number of memory failures experienced is unacceptably high, both ddr1 and ddr2, plus the odd memory controller issue. Mainlly 4gb but also 2gb dimms. Sometimes straight failure and deallocation, but also predictive errors quickly leading to deconfiguration and deconfig by assn.
And - what i s a memory diag error E3 ??
b) The cost of replacing memory is also very significant , given the number of errors we are getting (one a month) .
c) We note IBMs guidance to get the firmwar updated to cure as many memory bugs as possible - we have had a couple of crashes after memory errors which could have been prevented by fw update.
d) Then we query replacement method - quads,sets,pairs or 1 by 1.
General advice is to purchase replacement in identical quads, Info centre notes replace as a set (not seen a 'set' of memory so referred anywhere else ! ) I would expect pairs to be a good replacement (by memory bank - assuming identical ccin to other pair in quad), and IBM engineers expect to replace only whats necessary (1 by 1).
e) And then what's identical - by CCIN or by fru.

Just some experiences I thought I'd share that are causing me a pain at the moment !