Memory dividers - do they make a difference to performance?

[Saxamafone]

Just as the title says really!

I have heard a memory divider of 1:1 is optimum but I just wondered if it really makes that much of a difference?

[BUFF]

a difference, yes - a significant difference probably not.
& 1:1 isn't always optimal - depends on system.

[Blitzen]

Top question
Not many know...i dont and its handy to have some idea.

[GSte]

As Buff said 1:1 isn't always optimal, it does depend on the rest of the system.

The divider can be an issue sometimes.... on some boards I have found 4:5 to be unstable, and 2:3 on others. I'm not sure if that's due to the chipset or the particular board.

[Saxamafone]

What is the best way to check for stability? At the moment I have a very basic setup but my ram is overclocked from 667 to 700 with a 1:1 divider and I want to make sure it is totally stable.

[GSte]

Use Orthos and run a blend test:

Download Stress Prime 2004 BETA

Usually fails within half an hour if it's going to, although some people let it run for 4-8 hours to be sure. Personally I never need it to be that stable!

[Krumm]

however, I run my ram at 1 ghz, and the cpu at 3.5. I ran orthos on blend for 9 hours and it was completely stable, however as soon as i ran memtest, it picked up errors, so i had to up the voltage and loosen the timings to make it work properly.

[kalniel]

The answer to the question is, on modern computers, no.

Once upon a time memory controllers did incur a performance penalty in having to convert a non-1:1 ratio. Then AMD stuck the controller on their CPU and it was so fast as to make the problem irrelevant, then Intel improved both their northbridge and the way the chips used memory to also make the problem irrelevant.

Nowadays you will only see a very very small difference (all other things being equal) in ultra-sensitive synthetic benchmarks which have no bearing on real usage.

Instead, the flexibility of a non 1:1 ratio will often mean that you can improve the performance of something else (RAM/CPU) such that any minute performance hit from not running 1:1 is far outweighed by a real benefit such as higher CPU frequency.

[GSte]

however, I run my ram at 1 ghz, and the cpu at 3.5. I ran orthos on blend for 9 hours and it was completely stable, however as soon as i ran memtest, it picked up errors, so i had to up the voltage and loosen the timings to make it work properly.

Wow, never experienced anything like this personally.... usually Memtest doesn't find errors unless it's very unstable or the RAM is knackered. More than one testing program certainly doesn't hurt though....