So what's the deal with FB-DIMM?

Baraclese

Newcomer
I see this is only meant for servers/workstations at the moment. Are there plans to introduce this for Desktop PCs? I've read that you can increase the available bandwidth by plugging more FB-DIMM pairs into your motherboard. Are there benchmarks that show the benefits this brings?

*Someone gift me a MacPro please* :)
 
It's serial data transmission.

Think parallel ata vs. serial ata.

Costs are higher because a controller chip of sorts needs to be built into each stick.
 
I've read it adds a small bit of latency but will be very useful for increasing bandwith and capacity, and should end up being mainstream. Maybe with the AM3 socket ?
And FB-DIMM allows memory chip technology to evolve while retaining compatibility, i.e. you can have DDR2 or DDR3 FB-DIMM.
 
AFAICS, trading increased latency for increased bandwidth in this manner makes good sense when you have a heavily multithreaded/multi-CPU setup, but not when your setup has only 1 or 2 CPUs or threads (as witnessed by e.g. AMD's recent change from S939 to AM2, where a 100% increase in theoretical bandwidth translated into about 1-2% actual increased performance.)
 
it seems to me we can only go this route anyway, memory technologies keep increasing bandwith without doing much to latency (DDR, dual channel, DDR2, DDR3)
 
This presentation is a nice although a bit old introduction to FB-DIMM. Although FB-DIMM is very useful on servers, I don't think we need it on desktop computers, at least in near future.
 
Seems kind of like take one step forwards going from north bridge to on CPU memory controller (AMD) & then step back to north bridge but now stuck direct to the RAM :???:

[edit]Just read the above presentation & I see why its going to be good for servers but until we start having PC users wanting to use more than 4 DIMMs at once you have a pointless latency increase on the desktop (which would be why Intel & AMD are talking server only)[/edit]
 
Last edited by a moderator:
Back
Top