Odd multipliers

Odd multipliers impacting memory bandwidth

This is from TomsHardware. It was in their article on FX vs EE. This I found interesting, and would like to discuss:

The memory performance that we recorded for the Athlon64 FX shows that odd multipliers (12.5, 13.5, 14.5) slightly impact the memory bandwidth. The reason is the necessary and asynchronous operation of the memory, which leads to additional wait cycles.

Click to expand...

http://www6.tomshardware.com/cpu/20040419/cpu-scaling-05.html

On the link there are graphs where it shows a decrease in memory benchmarking on all the odd multipliers.

Not sure which board this should go in, general CPU or Athlon, since it deals with multipliers, but here goes.

Discuss!

Hmmmm

I am using 12.5 on my DFi Ultra Infinity B - wonder if that is slowing my memory down?

Might have to do some testing.

I've never heard of that before, and do'nt know why it would slow you down. Keep in mind people don't regard Tom's as being too reliable anymore.

Could be a problem with Sandra handling the half multis rather than anything else. But certainly worth further investigation using benchies other than Sandra to eliminate that as a possibility.

But how do the multis even effect the bandwith?

I've heard this for the last almost 5 years, and as a result, I avoid the half-multis like the plague. Generally, what was quoted from the article by the first poster is what the deal is; extra latency caused by switching between 12 and 13 clocks for a 12.5 multi. I would guess that DDR would either make this concern obsolete or would double its validity, meaning that one should avoid odd multipliers as well as halfs. I personally try to aim for powers of 2 with my multipliers, but this is not something I'm as crazy about as avoiding halfs.

The performance hit isn't that critical, but it's one more thing to factor in.