FSB : DRAM Guidence

Callum1991

Honorable
May 14, 2013
15
0
10,510
Hi guys,

As a little introduction, Im trying to explore the world of OC'ing a little with my recent budget build;

AMD A8 5600K APU
AMD Radeon 6670 1GB Dual Graphics iGPU
8GB Corsair Vengeance 1866 (2*4)
AS Rock FM2 A75M DGS MB
Along with some junk WD HDD.
Stock coolers

I'm looking at trying to safely eek a little more out of my system without frying it, basically its a bit of a learning curve for me.

What I can't see to get my head around at the moment is speeds and timings.

I've read that a 1:1 FSB:DRAM ratio is ideal, mine is 3:20 at the moment. I wouldn't even know if that is embarrassingly bad!

I don't think I'm getting anything like what i could out of my memory, CPUZ is telling me my DRAM frequency is 665.3.

I have to hold my hands up and say it is all a mystery to me! Things like CAS Latency is an absolute mystery to me!

I've had a play with OC'ing the CPU and GPU (3.6 to 3.9 and 810 to 900 respectivley), very modest as you can see.

Basically I am wondering if anyone could maybe clarify for me how all these speeds and timings are balance, or perhaps point me to a good beginners guide.


Many thanks in advance guys, happy new year!

Callum
 

bebop460

Distinguished
Jun 25, 2011
169
0
18,710
CPU-Z shows RAM freq. per stick. So if it's 665.3mhz X 2 sticks, you're running at 1330.6mhz, nowhere near your rating freq. of 1866.

First I'd run memtest to make sure the memory is still stable with your CPU overclocked, and immediately again each time you change the RAM config.

A RAM OC is usually not worth the hassle, with very little (if any) benefit, while putting added stress on the system and risking OS corruption (you probably want to back everything up before proceeding, to be safe). Also their are many different voltages that can need to be tweaked for stability. That said, I've heard AMD systems benefit more from a RAM OC than an Intel system. If you want to try it, I'd measure performance before and after and decide for yourself if it's worth it.

I learned from JackNaylorPE here http://www.tomshardware.com/answers/id-1795015/system-memory-overclocking-cpus-die-memory-controllers.html that you can calculate memory performance (theoretical nanosecond delay) by dividing the frequency by the latency:

"9-10-9-27........9/1866 x 1000 = 4.82 (1.50)/
11-11-11-27.....11/2133 x 1000 = 5.16 (1.50)"

In that case the 1866 memory is faster with only a 4.8 NS delay versus the 2133 delay of 5.1. If you do decide to proceed, I had success in looking at my own RAM spec on the mfr site, and all the spec.s of other higher performing modules in the same line.
-
For example, I have GSkill Ripjaws X ram 1600 (5ns delay). I looked at other higher performing Ripjaws X RAM, calculated their NS delay, and wrote down their spec. I tried 1866 C8 and failed. Tried C9 1.57V and succeeded, yet 2133 C9, 10 and 11 failed. In my case their was no notable performance gains in OCCT, 3DMark, Aida, or any other benchmarks.



 

Callum1991

Honorable
May 14, 2013
15
0
10,510
That's excellent thanks for the response!

I have tried a little over/underclock each side (1600/1066 Respectivley) but had stability issues. It may well be a voltage problem.

If I've got my sums right (which there is a good chance I haven't) my A8 is over clocked at 40x 100. That seems to run okay.

But as my bus speed is 100 and it's a quaddy, would I be right in thinking 800mhz ram speed would give me my 1:1?

4x100=800/2

In which case I would over clock and try for lower CL? Or not?

I'm unsure whether or not I would actually under clock my memory that far mind....

I think I could get Cl 7 on 800mhz.

So...

7/800=8.75
9/1366=6.58

The math tells me it's not worth doing.

Here's a screen dump of my memory(someone else's pic);

http://www.reboot.ro/reviews/review-corsair-vengeance-cmz8gx3m2a/
 

bebop460

Distinguished
Jun 25, 2011
169
0
18,710
Would that be 800mhz per module; 1600mhz effective frequency? Wouldn't the rated speed or faster (while considering NS delay) be better?

To be honest, I have no familiarity with FSB:DRAM Ratio whatsoever. You may want to post on overclock.net; great overclocking community there.

I would sooner try to get the maximum CPU overclock and stabilize the RAM. From there, you may wish to overclock the RAM and benchmark it to determine its benefit.

Downtuning the RAM seems like a waste of time to take a step backwards, but I know little of differences in AMD systems.

I have some notes for NS delay for different frequencies and latencies, should you want them.

An important note I should mention: in my experience, RAM modules seem to only be compatible with certain latencies. However, they could be overclock limitations on my modules only, but it seems like a heck of a coincidence. As I recall, GSkill listed it was only compatible with C9,8, and 7, and sure enough, all other latencies were unstable, despite all efforts to stabilize them. I'd look up your model on Corsair's site and see if they list compatible latencies and use it as a ref. if you decide to OC the RAM.
 

Callum1991

Honorable
May 14, 2013
15
0
10,510
Thats right!

I've since O/C'd it to 1600 and its stable (had to give the NB abit more voltage).

Corsair basically say its good for;

1333 @ 9-9-9-24
or;
1866 @ 9-10-9-24

Im now 1:8 in terms of ratios.

I loosened the timings and i've got it stable (And beating the benchmarks, I added 2 points to my RAM score for windows experience index, however useful that is!)

Thanks for all your help bebop, now I shall endeavour to find out how matching my GPU/CPU frequencies should be done.

I get the impression the best answer with OCing isn't always 'turn everything up to 11'.

Many thanks!
 

bebop460

Distinguished
Jun 25, 2011
169
0
18,710
Happy to help. I'm glad you got it configured to your liking. Did you run memtest ~8-12hrs to make sure it's stable?

I'd offer to help with matching GPU/CPU freq.s, but I'm unfamiliar with it.

Yeah, oftentimes it seems like OCing eventually reaches a point where either it hits the wall with temperature, or massive extra stress on hardware for miniscule performance gains.