Loren said:
I've encountered multiple machines that failed Memtest after working
for an extended period.
As for the heat sinks--could someone have been overclocking?
Manufacturers use heat spreaders on DDR2 and DDR3, even when they
aren't needed.
The last Ballistix I bought, had heat spreaders. If you were to
remove the heat spreader, instead of seeing a Micron branded chip,
you'll see a custom logo there instead, so you don't really know who
made the chips on it. You'd assume they were Micron, but Crucial in
recent years, has been known to use a variety of parts as circumstances
dictate.
Non-enthusiast RAM, tends to come without heat spreaders.
Whether a heat spreader is an advantage, depends on the overall
dimensions of the module that result. In some cases, a large heat
spreader, blocks the air channels between DIMMs, so the DIMMs in
the center of the DIMM "sandwich" can run hotter than they should.
In fact, if you use four modules, they may run cooler if they come
without heat spreaders in place. It leaves a little room for
air channels.
Some goofy module designs, even include "fingers" that stick up
above the module. That is intended to compensate for the poor
air channel performance. And then the fingers can bump into other
objects which are just above the DIMM socket area.
The only technology, where spreaders were essential, was RDRAM/RAMBUS.
That's because, continued access of a single memory location
(or localized access) would cause just one chip on the module
to heat up. An article I read, claimed before RDRAM came out, that
the chip could have up to a 4W power dissipation. By comparison,
a modern DIMM, with shrunk geometry chips on it, now runs 2W for
the entire module. So RDRAM had heat spreaders riveted to the module,
to ensure the "hot spot" problem was solved. DDR/DDR2/DDR3 modules
don't have the same hot spot issue, because they're a parallel
technology. The entire bank heats up together, and with less heat
from each chip.
DDR/DDR2/DDR3 modules can heat up, if you do enough of the right kind
of cycles on them, and some Northbridges actually had provision for a
feature to detect excessive, power hungry type cycles, and reduce
the access rate. It was considered better to do it that way, than to
glue a thermistor to the module and check for abnormal temperature that
way (to implement throttle on overheat).
DDR module power ratings, are based on an "industry standard cycle mix",
which assumes a less than punishing access pattern. If a person was
crafty, say a hardware designer doing a custom design with RAM for
storage, they could arrange to do long bursts of reads or writes
(i.e. 256 locations in a row), and the power dissipation that results,
would end up being higher than the spec sheet value (because the spec
sheet assumes a more relaxed pattern). For comparison, processor
accesses are done in terms of cache line fill/evict (burst of 4 or 8 perhaps),
which doesn't take the form of a burst of that length. The bursts are
shorter, with some "dead cycles" in between.
Paul