tiny disadvantage of registered RAM?

[Nomen Nescio]

I recall several years ago a colleague saying he would not
use registered RAM because it adds one cycle to the latency.
In those days, I think CL3 was the standard timing.
But now we have much faster RAM with the latency gone up
to 9. So an extra cycle 10 versus 9 is not worth SFA
is it?
So if you want an "engineering workstation", go for a
server mainboard with registered?

 

[Paul]

I recall several years ago a colleague saying he would not
use registered RAM because it adds one cycle to the latency.
In those days, I think CL3 was the standard timing.
But now we have much faster RAM with the latency gone up
to 9. So an extra cycle 10 versus 9 is not worth SFA
is it?
So if you want an "engineering workstation", go for a
server mainboard with registered?

This sounds like a reasonable analysis.

The registered though, could be eating into the upper
speed limit. So instead of DDR3-2100, you're limited to
some lesser number. It all depends on how well the
register chips can keep up with the ever-advancing
clock speed. A registered DIMM never seems to be available
at as high a speed as an unbuffered.

For users who simply cannot afford memory errors, they
probably don't have a choice in what they end up with
for a memory subsystem. You want "reliability", expect
some small impact on performance. With ECC or ChipKill
in a system, you stand a better chance of knowing whether
your results are potentially corrupted or not.

Paul