Sometimes less is more
With regard to memory "speed" it is a case of swings and roundabouts. If one looks at increasing the DRAM frequency at some point one has to increase the timings simply because of the DRAM latencies which have - individually - a minimum below which one just cannot go.
Another thing to consider - which I almost never see mentioned - is the actual distances on the motherboard the signals have to travel.
Signals have to travel the distance and return within a very narrow time frame and those signals have to reach the components and return in the right order. One main factor influencing this is distance. Another factor is the conductivity of the material and how that material's conductivity is influenced by temperature.
The effects of this are minuscule - until one starts increasing the frequency of the usage of a bus. At some point a limit is reached. A motherboard might be specified for a certain "speed" but this assumes that everything during the production process has gone absolutely perfectly - differences to be measured in atoms can make the difference between yea and nay.
To increase the DRAM frequency beyond a certain point one has to increase the voltage and of course that just introduces more heat and it might not be enough to compensate for the other factors.
So with interacting computer components there comes a point where the "should be able to" intersects with the "will be able to".
No one component is to blame, it is the interaction of the components which has to be pretty close to perfect for the "should" and the "will" to coincide for the higher performance to be realised.
Scottie's famous admonition to Kirk should be borne in mind, "Ye canna change the laws of physics".
The DRAM timings are plain and simply instructions for components to do NOTHING for a certain period of time before they try to send a signal again. This is why I talked about swings and roundabouts with regard to "speed" at the beginning of the post.
Let me put this another way. Over thirty years ago I worked in a concrete factory doing piecework - I was paid according to how much I produced. Although, due in no small part to having had two spine operations, my body has gone to hell in a hand-basket at that time I was very fit, fast and strong. I got to a stage where I was earning a lot of money - however I also crossed a tax threshold where I was actually coming out with LESS money in my pocket than if I produced less and concomitantly got less gross pay to tax.
Consider this in relation to increasing the DRAM frequency - your pay - and the DRAM timings - your tax. At some point the tax you have to pay in timings exceeds the pay you get in increasing the DRAM frequency.
My advice is, forget what the rig you have should be able to do. Start off at the lowest DRAM frequency rating with the motherboard settings set to "auto" and gradually raise it seeing how that affects the adjustment the board makes to accommodate the change.
Forget "XMP profiles" which are based on a best case scenario; however, as I tried to show above, the assumptions made with regard to the rest of the hardware to actually be able to live up to them is outwith the control the engineers responsible for the RAM modules who create those profiles.
Sure you might get the DRAM to work with the higher frequency settings (your pay) but the price which is exacted with regard to the timings (your tax), i.e. the amount of time your memory subsystem spends doing nothing, could actually end up with you getting LESS performance.
Last edited by Nec_V20; 01-05-13 at 04:41.