Clocking RAM to the right speed.

[vntg]

So for a 1066 FSB, theoretically DD2-533 in dual channel is all that is needed to max out the bandwidth of the fsb.

My question is, which of these cases would yeild better memory performance in practice?

If my FSB is 1066 and I have DD2-800, should I..
1) run my ram 1:1 at 266Mhz CL4?
Or would my system be faster if I..
2) run the ram async at 400Mhz CL6?

Im a bit confused, would there be any performance gains or any point to running the RAM faster than the base clock?

..and please correct me if i am wrong in anything i said above.

 

[saint19]

Hi.

You can run at 400MHz CL6, but you will not see an improve on performance 'cause the CL is more high (CL4 agains CL6).

 

[jezzarisky]

So for a 1066 FSB, theoretically DD2-533 in dual channel is all that is needed to max out the bandwidth of the fsb.

My question is, which of these cases would yeild better memory performance in practice?

If my FSB is 1066 and I have DD2-800, should I..
1) run my ram 1:1 at 266Mhz CL4?
Or would my system be faster if I..
2) run the ram async at 400Mhz CL6?

Im a bit confused, would there be any performance gains or any point to running the RAM faster than the base clock?

..and please correct me if i am wrong in anything i said above.

If you're running your CPU on base clock, than you would see very little benefits from OCing the ram to 400Mhz, and it would best at 266Mhz @ CL4 of course if you can clock the timings down lower, or keeping the CL4 and push the bandwidth up more that works too (of course you probably won't see any super noticeable benefits from this).

 

[vntg]

So, if my FSB is 1066, there would still be benefit if I run my RAM at 400Mhz CL4 (lets pretend its possible) vs 266Mhz CL4? At 400Mhz, the bandwidth of my RAM is far larger than my FSB, what would be the point if my FSB can't handle that much data anyways?

 

[jsc]

Definition time (attention purists, I'm talking about DDR2 and I'm simplifying a little ):

Core2 CPU's use a frontside bus (FSB). The FSB is a thing with two main characteristics: speed which is usually defined in MHz and width which in the Core2's is 64 bits wide. We are concerned with the speed.

Using the Q6600 as an example, the FSB frequency is 266 MHz. The matching DDR2 memory clock for that frequency is 533 MHz (266 X 2). DDR2 memory transfers two chunks of data for each bus cycle, hence double the frequency. So, to run 1:1 at an FSB of 266 MHz, we need DDR2-533 RAM. What CPUZ does is a little confusing. It will tell you that the memory frequency is 266 MHz for a 1:1 ratio. Your BIOS will tell you that at that speed, your memory clock is 533 MHz.

The FSB clock is 1066 MHz (266 X 4). The bus is "quad pumped". It transfers 4 chunks of data into and out of the CPU each cycle. So each FSB cycle generates 4 FSB clocks.

Now, if you increase the FSB frequency to 333 MHz, the corresponding memory clock is 667 MHz and the FSB clock is 1333 MHz.

Running the memory faster than the FSB only means that the memory has to sit and wait for the CPU to catch up.

You will get more performance by running the FSB:RAM ratio at 1:1 and tightening the memory timings than trying to overclock the memory. Case in point: my first Core2 system - E6600, eVGA 680 motherboard, 2 GB DDR2-1066 Crucial RAM, ACF7P cooler.

I was running the CPU at 3.3 GHz (367 MHz X 9). I spent a lot of time working my memory timing down from 5-5-5-15-2T to 3-3-3-7-1T. 😱 Memory i/o increased about 7%. Unfortunately the increase in system performance was unnoticeable. 🙁

We talk about overclocking memory here:
http://www.tomshardware.com/forum/251715-29-ratio-myth
----------
Overclocking since 1978 - Z80 (TRS-80) from 1.77 MHz to 2.01 MHz