Memory Questions

[Liquid_Static]

Hello everyone, I've been a lurker on the forums for a while, but decided today I'd like to participate in some of the amazing discussions these forums have to offer.
ANYWAYS...
I have a question about overclocking memory, I have a 16GB set of DDR3 rated for 9-9-9-24 at 1.5v 1333 MHz, however i can set the timings to 10-10-10-27 at 1.5v 1600 MHz. Is that faster? In short, I'd like to know how to take memory latencies into account when measuring speed.

 

[Idontcare]

Welcome to the forums Liquid_Static :thumbsup:

Hello everyone, I've been a lurker on the forums for a while, but decided today I'd like to participate in some of the amazing discussions these forums have to offer.
ANYWAYS...
I have a question about overclocking memory, I have a 16GB set of DDR3 rated for 9-9-9-24 at 1.5v 1333 MHz, however i can set the timings to 10-10-10-27 at 1.5v 1600 MHz. Is that faster? In short, I'd like to know how to take memory latencies into account when measuring speed.

First you have to normalize the latency. 9 latency at DDR3-1333 is not the same as 9 latency (or 10) at DDR3-1600.

A 10 latency at DDR3-1600 is the same as a (1333/1600)*10 = 8.33 latency at DDR3-1333 speeds.

So the reality is, for your specific setup, not only is your ram bandwidth higher at DDR3-1600 but your latency is actually improved as well because you did not relax the latency timings as much as you increased the ram's speed.

That said, the question is "how would you measure the performance impact?" - and the answer there is you can use synthetic benchmarks like Aida64 or SiSoft Sandra but understand the real-world impact in the apps you actually use will be practically negligible.

 

[Liquid_Static]

Thank you! That was extremely helpful.

 

[Cogman]

It should be noted that, for 99% of applications, memory speed simply doesn't matter. It is fast enough. The rare exceptions are those applications that deal with huge amounts of data (File compression, video encoding, etc). And even then, it generally isn't that big of a difference.

The only other place I can think of where memory bandwidth might play a larger role is in an environment that is doing a lot of context switching (for whatever reason). In that case, the cpu might be swapping out memory frequently enough to make a noticeable difference in application speed (it would be hard to measure). For the context switching thing, think of a computer that is running 500 different applications at the same time each requiring a fair amount of CPU and memory resources.