3-4-3-9 DDR2 800 or 5-5-5-15 DDR2 1066?

[chriswarner]

I guess my question is whether timings or bandwidth is the more important factor in performance in an Intel chip/motherboard. I have been told in other posts in these forums that it makes little difference, as Intel chips do a great deal of on-chip caching. And my own meager tests in the last few days (comparing 4-4-4-12 800 to 5-5-5-15 1066) seem to support this. But I thought I'd pose the question to the bigger, smarter group.

 

[CompuTronix]

The following explanation is a re-post from the thread "who uses a 1:1 ram ratio?" - http://www.tomshardware.com/forum/247906-29-ratio#t1765557

CompuTronix wrote:

"...Ah, but wait... there's more! Let's consider some of the additional variables involved in the memory big picture:

A ratio of 1:1 provides the best level of stability, since the memory controller, which is an integral part of the northbridge chipset for Intel processors, does not need to translate data flow across the FSB between the memory modules and the processor(s). Also, since memory and processor FSB clocks are synchronous at 1:1, (400:400 or DDR 800), there is no additional latency introduced.

If a minimal ratio of 4:5 (400:500 or DDR 1000) is used, then the resulting increase in memory frequency is effectivey cancelled out by the latency introduced in translation across the FSB between memory and processor clocks, and no increase in memory performance can be noticably detected in benchmarks. Also, asynchronous or mismatched clocks, create an element of potential instability within the memory controller, so depending on the chipset, an increase in northbridge and memory voltage is required for stability, which results in more heat, and less FSB overclock ceiling.

If a more aggressive ratio of 2:3 (400:600 or DDR 1200) is used, then the increase in memory frequency can marginally overcome the latency introduced in translation across the FSB between memory and processor clocks, resulting in a marginal increase in memory performance, which typically yields an increase in memory benchmarks of 2 to 3%, and is relatively negligible in terms of overall system performance.

In the case of expensive DDR3, where a ratio of 1:2 (400:800 or DDR 1600) or 2:5 (400:1000 or DDR 2000) is used, even with the tightest timings, an increase in memory benchmarks of only 3 to 4% is yielded over DDR2 800, which once again, is relatively negligible in terms of overall system performance. It's also noteworthy to consider that memory timings of 4-4-4-12 compared to 5-5-5-15, will yield an increase in memory benchmarks of less than 2%. However, for those of us who have the need for speed, we'll take whatever we can tweak.

Additionally, at equal specifications, 4 memory modules offer less FSB overclock ceiling than 2 modules, since more northbridge and memory voltage is required to maintain stability, and 4 slots require twice the current of 2 slots, again resulting in more heat, which typically is pulled into the CPU cooler, where it can increase processor temperatures by a few unwanted degrees.

I hope this helps to provides a greater degree of perspective (no pun intended).

Comp

Incidentally, I run my rig at 1:1..."

 

[chriswarner]

As always, CompuTronix, this is very helpful. I have only one question. You stated '4 memory modules offer less FSB overclock ceiling than 2 modules'. To my noob mind, I can read that both positvely and negatively. So, can 'FSB overclock ceiling' be restated as '2 memory modules can be run at an overall faster frequency than 4 modules'?

 

[outlw6669]

C3 DDR2 800 will allow you to access your data in 7.5ns vs around 10ns for C5 DDR2 1066.
You also have the minor timings lowerd which gives you a further boost in access time.

 

[Guest]

I'd go with the faster DDR2-800.

 

[Scotteq]

Short answer - The DDR2 800 with the nice tight timings.

Long answer?

As always, CompuTronix, this is very helpful. I have only one question. You stated '4 memory modules offer less FSB overclock ceiling than 2 modules'. To my noob mind, I can read that both positvely and negatively. So, can 'FSB overclock ceiling' be restated as '2 memory modules can be run at an overall faster frequency than 4 modules'?

Certainly, there is the potential for two to be run faster, yes. And if you're going for benchmark numbers/the highest possible overclock, then you should certainly stick to two DIMMS. As Comp pointed out, running 4 modules is more stressful on your mobo, and you may be forced to choose one (fast) or the other (large quantity) in order to achieve a stable system. Given DDR2 prices, and the gains involved in using fast RAM on a FSB based system?? More is better.

And as we all know, all motherboards/brands are not all created equal - even down to the individual level since manufacturing tolerances do produce supposedly 'identical' boards with slightly different capabilities. This is (part of the reason) why you may have some individuals with brand #1 saying it's the greatest thing ever, and someone else with the same exact setup who is unable to achieve similar results. Maybe person A's board got a set of well matched and top of the spec range capacitors on his board's power regulation subsystem, while person B wasn't quite as lucky.



So, having said that: Are you going for benchmark numbers? If not, and you have a quality motherboard, then IMHO there's no reason not to run 4 modules provided you're prepared to manually set you system up and perhaps do a little extra tweaking. To my mind, the real question is "How much memory do you feel you need?" Add a little to that, and see what modules you need to install to get there. i.e. If you feel 2GB is probably enough, then maybe you'd be best off thinking about 4GB. You aren't sure your board will run 4x1GB DIMMS well? Go 2x2. Regarding speed and timings, we already know that on an Intel/FSB system it doesn't make a whole lot of difference. So make sure you have enough, rated fast "enough" for your intended overclock, and move on to the next subsystem.


If anyone cares, and only intended as an example - My own box has a stupidly large amount of RAM, for no particular reason other than "Because I Can". But I'm not going for a huge overclock, either: 3.2 Ghz using an 8 multi on a 400 base (1600Mhz) Front side buss. The motherboard I chose currently runs 2GB x 4, at 1:1, 4-4-4-12 2T timings at the RAM's rated voltage. Now - When I had DDR2 1066 modules in there - It could overclock a pair of 1 GB sticks even though the mobo is 'only' rated for DDR2 800. But it wouldn't run 4 x 1 at 1066. At least with my rather limited skills, the best I got at rated memory voltages was DDR2 1000. I could get the 1066, but not without pushing stupidly hard. In the end I decided the e-peen wasn't worth hosing up my computer, so I downclocked to 800. Performance wise, I noticed no difference. Temps dropped by almost 10 degrees, and it was 24 hour P95 Blend stable as soon as I made that change. Clearly time to declare victory.

In this case, the stress of overclocked memory plus the stress of 4 modules was more than the board I chose was willing to put up with. But 4 modules of twice the capacity, at a non-overclocked speed? No Problem! This is why I ask what you have in mind. If you want to bang out benchmark numbers, then definitely stick to 2 fast DIMMS. Otherwise, IMHO, it doesn't matter much as far as every day usage. And therefore you'd be best off worrying about your processor overclock and not breaking any memory 'rules' in the quest for performance.


The above may change with Nehalem, and current AMD based systems do benefit more from faster memory more an Intel box. But for the time being, anyways? Bigger is Better!