Something I have rarely seen talked about until recently is using Half-Multipliers when Overclocking.

For people with Intel CPU's and locked Athlon XP's this will not pertain to you, but you might have a friend with one you could pass the information on too.

For those with Unlocked Athlon XP's, Duron's and AMD 64's...this is something you should consider when Overclocking them.

What I am talking about is that most people have options in there Bios to allow using Half-Step Multipliers (.5) and it may not be giving you all the performance you thought you were getting.

Lets compare a few situations, assuming you have an Unlocked Barton 2500+ and an Abit NF-7 for example:

If your goal is to run at aproximately 2400mhz (2.4ghz) you might set it up as 200x12=2400mhz, or you could also run it at 209x11.5=2403mhz to get a little bit more memory bandwidth, since that should improve your scores in benchmarks right?

From what I have been reading about, the second situation may not be so ideal after all. It seems that there is a bit of controversy about this, some articles I have read say that using 209x11.5 would actually not be running the memory at 209FSB, because the "Half-Multipliers" do not actually work properly.

AMD says it does not officialy support this function, even though it is commonly available, and I see people use them at times, myself included.

It all began when I began noticing that I could run my memory at higher speeds with 8.5x instead of 8x on my Sempron 3100+.
I also was getting some odd results when looking at Clock Gen when I used a 9/10 memory divider (Another subject, but might be related) so I thought I would check it out a bit closer.

I ran across an article HERE, wich says that the motherboard they tested was actually "Jacking with the system bus" and "Faking" the results.

Although I am not sure if I completely agree with his results, I think he is on to something, and has taken the time to try and document it software like Everest and AIDIA 32. He believes that the FSB speed is actually running about 5% faster than what you are setting it for, so that 209 FSB might actually be running at 219FSB!

That can be a real problem if your memory won't run at those speeds, however there also have been several threads in other Forums (But not as well documented) that suggest it may be running slower than 209 FSB. It would actually be more like having it at set at 199FSB.

I have been trying to figure this out by doing some testing, and have come up with some pretty inconclusive results at this time, I would like to hear if anyone else has noticed anything "Odd" about results when using the Half Multipliers on there systems

Sorry i can't help you with any feedback on the multiplier issue, but at least now i have a better understanding of how they work....thanks

Dang good article! Keep them coming.

You know, it's funny you mention that, I thought I could run the bus speed of the memory asynchronously from the FSB....

Of course I haven't done extreme Unaclocking in a while so I may be wrong.

i believe i talked to you on msn about this a while back when i was trying to find my max OC with my ram.

when i ran the 12.5 multi i could get 2.85 ghz but the 12 at 2750 mhz would run faster in ALL benchies. A full 100 mhz diff and the fsb was the same. Both were SOLID overclocks and primed for a day each.

Benchmarks showed the 12.5 multi as being the equivalent of my 2600 mhz oc.

So i guess in my case the half multi's are ****. NF7-s and Corsair xms 3200 by the way. 2500+ mobile

Yes, they work.

Yes, they work as BA said, the original point of this thread was to discover if they are running the memory at the correct speed.

The basic question was it might be underclocking/overclocking the memory slightly when using half multipiers, some odd non linear scaling issues were noted in the articles I mentioned.
I was trying to find out what was going on, and see if anyone here had done any homework and or testing on this subject.

From an Overclockers standpoint, there could be performance advantages to having this option avaialble, just another tool, similar to using a mem divider/multiplier.

If you are unaware of it, it could be costing you performance, if this theory can be proven one way or another.

When time allows, I would like to investigate this further and see for myself, I just noticed it was odd on both my AMD 64 and MobileXP..it seemed that I could push a higher FSB when using a half mutiplier than with a whole numbered one.

Using a whole multiier I could not reach 300MHZ FSB 1 to 1 (7x300=2.1ghz failed me test) it would start becoming unstable no matter what settings I tweeked (2T instead of 1t, loose memory timings, more voltage, looser timings etc in the 285-300FSB range.

When using a half multiplier lower (6.5x 300 passed mem test) or higher (7.5x300 passed mem test) it would run stable.

Further testing with that same board using a 3/4 memory divider allowed me to reach the limits of either the board itself, or the on die memory controller at 320FSB.

I honestly don't know if the Crucial Balllistix PC3200 I had was realy running at a true 300FSB (DDR600) or not when using the half multipliers


I found the best overall performance for my combo was at 275x9 (2475MHZ) CAS 3-3-3-10 with CPC on (1T) and HTT at 3x275 (825MHZ)

FYI I also tried running the HTT (LDT in the DFI Bios) at 4x275 (1100MHZ) and found it would not boot up, AMD-64 does not like HTT anything much over 1000MHZ from everything I have read so far, and there is no performance penalty of note while running under 1000MHZ either.