RAM vs CPU performance

[bloodandsoil]

Which would be better (gaming mostly):

A) 1:1 ratio
FSB = 333MHz
333 * 9 = 2997MHz CPU clock
333 * 2 = 666MHz RAM clock

B) 5:4 ratio
FSB = 320MHz
320 * 9 = 2880MHz CPU clock
320 * 1.25 * 2 = 800MHz RAM clock

Edited to add:

CPU = Core 2 Duo 2.4GHz (running at stock defaults)
Mobo = Asus P5B-Deluxe
RAM = Crucial Ballistix DDR2-800

 

[IcY18]

A

 

[Scougs]

I would suggest trying one of these solutions:

ratio 1:1

FSB 400
CPU 8*400 = 3200
RAM 2*400 = 800

or

ratio 1:1

FSB 400
CPU 7*400 = 2800
RAM 2*400 = 800

 

[SuperFly03]

Go with scougs advise.

Stay away from 5:4, bad bad bad bad. lol.

Here is why

Either go 3:2 (even worse than 5:4 when OC'ing) or 1:1.

 

[bloodandsoil]

What's the diff between these two:

1:1 ratio
FSB = 400
CPU 7 * 400 = 2800
RAM 2 * 400 = 800

5:4 ratio
FSB = 320
CPU 9 * 320 = 2880
RAM 2 * 1.25 * 320 = 800

They both have the same effective CPU and RAM clockspeeds...

 

[SuperFly03]

What's the diff between these two:

1:1 ratio
FSB = 400
CPU 7 * 400 = 2800
RAM 2 * 400 = 800

5:4 ratio
FSB = 320
CPU 9 * 320 = 2880
RAM 2 * 1.25 * 320 = 800

They both have the same effective CPU and RAM clockspeeds...

Read the article I linked... it explains everything.

 

[Scougs]

What's the diff between these two:

1:1 ratio
FSB = 400
CPU 7 * 400 = 2800
RAM 2 * 400 = 800

5:4 ratio
FSB = 320
CPU 9 * 320 = 2880
RAM 2 * 1.25 * 320 = 800

They both have the same effective CPU and RAM clockspeeds...

The difference is that with 1:1, the CPU can actually take advantage of the full memory speed because the FSB will be capable of matching theoretical bandwidth with the memory.

 

[bloodandsoil]

The difference is that with 1:1, the CPU can actually take advantage of the full memory speed because the FSB will be capable of matching theoretical bandwidth with the memory.

What do you mean by the FSB matching theoretical bandwidth with the memory?

 

[Scougs]

Generally you get optimal performance when the memory clock and the fsb clock are the same. There are two reasons for this:

1. Minimal memory access latency - When the CPU needs to access memory, it doesn't have to wait for the next clock of the memory to get started because they are sychronized.

2. The theoretical maximum data rate of the FSB and the memory are matched if the memory is running in dual channel mode and the clock speeds are matched. Thus, neither of them pesents a bottleneck in theory. In actuallity, one may be a little more efficient than the other.

 

[bloodandsoil]

So according to what you said, this would be better:

1:1 ratio
CPU 266 * 9 = 2394
RAM 266 * 2 = 532

Than this:

3:2 ratio
CPU 266 * 9= 2394
RAM 266 * 2 * 1.5 = 798

Even though the second one has a higher RAM clock? The first one is better because the CPU and RAM have matching clocks?

 

[unclepauly]

Go with scougs advise.

Stay away from 5:4, bad bad bad bad. lol.

Here is why

Either go 3:2 (even worse than 5:4 when OC'ing) or 1:1.

Here is the conclusion to your article. "So does Core 2 need high speed memory to shine? The answer is a resounding no!"

After taking out all other system bottlenecks memory performance only gave 5% increase on avg. from ddr2 400 up to ddr2 800. Not what I call anything substantial.

 

[mitch074]

It is this 'average' 5% that bothers me; depending on what your main activity is, careful timing would mean a 2-20% increase.
What can be read from the article though is that very low timings have more impact than high speeds. So, I'd recommend A:
- higher CPU speed (which is indeed the most important thing in this system)
- sync'ed FSB and RAM speed (lower latency)
- try lowering the RAM's timings manually if its SPD doesn't include timings for this clock speed (a 5-5-5-15 DDR2-800 module should be able to handle 4-4-4-12 at 666 MHz easily. If at all possible, try to test 3-3-3-8 timings, but cool your RAM well if overvolting and cross your fingers)
SPD timings are often a bit conservative; don't hesitate to experiment (however, be ready to reset your BIOS often).

 

[SuperFly03]

Honestly, there is no real reason to cool your ram until you get above 2.8-2.9v. My DDR is running at 2.85v and is still cool to the touch. 2.8v would be quite high for DDR2, but I doubt the heat output would require active cooling.

The conclusion of the article is that 5:4 is a terrible divider to use and only once you reach 3:2 does the increase in speed make up for the increase in latency from running the FSB/MEM out of sync. So if you are overclocking you assume a 1:1 ratio and buy ram according to what you want to end up.

 

[Scougs]

Honestly, there is no real reason to cool your ram until you get above 2.8-2.9v. My DDR is running at 2.85v and is still cool to the touch. 2.8v would be quite high for DDR2, but I doubt the heat output would require active cooling.

Um, don't be so sure. DDR2 runs somewhere in the neighborhood of 25-100% higher clock speed than DDR memory, and in extreme cases even higher. All those extra clocks add to power consumption. At 2.9v, you would probably kill your memory almost instantly. Keep in mind that the officailly specified voltage for DDR2 is 1.8v if I'm not mistaken.

 

[mitch074]

@Scougs: SuperFly03 is talkint about DDR and not DDR2. DDR2 runs at a lower voltage but at twice DDR's frequency - as such it actually uses up much more energy than DDR and heats up much more easily.

 

[Scougs]

@Scougs: SuperFly03 is talkint about DDR and not DDR2. DDR2 runs at a lower voltage but at twice DDR's frequency - as such it actually uses up much more energy than DDR and heats up much more easily.

I realized that he was talking about DDR, but it sounded like he was projecting what may be true about DDR to DDR2 in regard to voltage.

 

[SuperFly03]

Oh no, I was illustrating that DDR runs cool at 2.9V @ 250MHz and my DDR2 runs cool with 2.2v @ 503MHz. Yes, DDR2 Spec voltage is 1.8v, which is almost 40% lower voltage than DDR. Basically don't jump on the active cooling so quickly, it is likely that natural airflow from the CPU/Chipset can cool the RAM adequately.

DDR2 at 2.9v would probably not even boot, if it did, with any activity it would lock up or fry, you are right.

 

[mitch074]

I was mentioning RAM cooling as in 'be sure there's sone airflow on it' - in cramped boxes, the RAM can end up lost in cables and outside the system's airflow.
As such, either ensure that most cables are tucked away, that the airflow between the front case and the PSU is not too indirect, an that the drives don't heat up too much.

Now if you can't do that, get RAM heat spreaders and try to orient some case airflow towards the RAM.

Actively cooling the RAM is overkill.

 

[SuperFly03]

Actively cooling the RAM is overkill.

So is the 70mm screamer I put on my NB... but some of us do it anyways... :twisted:

 

[celewign]

This is kinda like math. Grrrr.
-cm

 

[KeithCash]

B


Cheers

 

[mitch074]

in this case, where there is no real overclocking, it's not - in a well cooled system. If you have to actively cool your RAM on a system you're barely pushing, then there's a problem somewhere.

 

[mitch074]

agreed - it's just that the system at hand is not overclocked. I the question was about an O/C'ed system, you'd have a valid point. I was merely pointing out that you're a bit off-topic.
Moreover, it is risky to mention active cooling of the RAM in this situation, because if it gets too hot, then there's a hot spot somewhere in the system -and another, more pressing problem to solve.