A64 non-integer CPU multipliers: useless? Then explain this to me...

Nearly every A64 overclocking guide asserts that non-integer CPU multipliers (eg. 8.5, 9.5) are useless, because of the way the memory dividers are assigned for the A64 architecture. However, I am a bit confused now, because I believe I have stumbled upon a situation in which a CPU multi of 9.5 actually provides me with a memory divider that is unattainable with any other combination on my motherboard.

The memory divider options on my motherboard are as follows:
[2:1 DDR400] -- 1:1 mem / HTT ratio
[5:3 DDR333] -- ~5:6 mem / HTT ratio
[3:2] -- ~3:4 mem / HTT ratio
[4:3] -- ~2:3 mem / HTT ratio

Because my memory overclocks so poorly on my motherboard, I really can't get the most out of my Venice CPU with the dividers that are available to me and using integer CPU multipliers. I really need to keep the memory under 210 MHz, regardless of timings, to achieve stability. Let's assume I want to run my CPU around 2600 MHz. If I run 260x10 with the [5:3 DDR333] divider, or 288x9 with the [3:2] divider, I get a memory speed of 216.7 MHz, which my system cannot handle for whatever reason. What I really need is a CPU/13 setting. How can I get this? According to the equation given in all the overclocking guides, the divider is calculated as follows:

CEIL(CPU multi/(mem:HTT ratio))

Using an integer CPU multi, there is no way that this equation yields 13, at least with the mem:HTT ratios available on my board. However, if one uses 8.5 or 9.5 for CPU multi in the equation, 13 is achievable. For instance,

CEIL(9.5/(3/4)) = 13

With these settings, I could theoretically run 273x9.5 = 2594 MHz, while keeping my memory at 200 MHz! This seems like a perfect solution, except for the fact that Asus' BIOS support sucks and my board doesn't have a working [3:2] divider at the moment.

Anyhow, I'm asking knowledgeable overclockers out there to enlighten me if I'm completely missing something here. Where have I gone wrong in my thought process, or is this a real example that contradicts the notion that non-integer CPU multipliers are useless on A64 systems? Thanks in advance!

In your (rare) case, yes, it does appear that the 9.5 gives you one extra option...however, on nForces, there's a 143 MHz ratio that when coupled with the 9.0x multiplier would yield the same CPU/13 divider.

Gautam said:

In your (rare) case, yes, it does appear that the 9.5 gives you one extra option...however, on nForces, there's a 143 MHz ratio that when coupled with the 9.0x multiplier would yield the same CPU/13 divider.

Click to expand...

Yeah, I agree that my situation is pretty rare. I may flash back to an earlier BIOS version to see if the 9.5 multi paired with the [3:2] setting gives me a CPU/13 divider. I still think I may prefer to run 245x10 with memory at 206, rather than 273x9.5 with memory at 200. I would need more Vcore for the latter option, and performance is more than enough with my current settings.

not too rare, my friends pc i built, we used 319 x 8.5, because it ran everthing faster than 300x9

when using a x.5, when it divides for the memory, it rounds up for the divisor

so in his case, 319 x 8.5 divided by 9, so memory ran at 301.xxxxx, on air