Archived from groups: alt.comp.hardware.overclocking (
More info?)
Spajky wrote:
> On Thu, 23 Jun 2005 18:17:13 +1000, "Michael Brown"
> <see@signature.below> wrote:
>
>> Spajky wrote:
>> [...]
>>> ALL frequency versions of A64 have a system bus clock @
>>> 2GHz & default FSB/mem divider @ 10 !!!!
>
> forgot to mention FX series (1,6GHz & 8 ; same as Semprons)
All K8 CPUs have a system clock speed of 200MHz. Socket 939 CPUs have a 5x
hypertransport multiplier by default (resulting in a DDR'd 1GHz bus), socket
754's have a 4x hypertransport multiplier (giving a DDR'd 800MHz bus), and
socket 940's come in both flavours. Nothing stopping you besides BIOS
limitations (and obviously overall chipset hypertransport limitations) from
running it at multipliers of 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 4.0, or 5.0
though.
>> Multiplier for core clock
>>> frequency is another stuff !
>>
>> Incorrect on all three counts. Read
>>
http://www.emboss.co.nz/amdmults/k8mults.html
>> This is documented in AMD tech docs and has been widely tested and
>> found to be true (hence the discovery of the 5 exceptions).
>
> yeah, I read that, & even the author is not 100% sure & I already know
> that Oscar Wu's table (& some others too!)
>
> ... There was/is a lot a mess about finding true memory clock on "k8"
> systems,
There's a lot of mess about the actual memory clock out there for three
reasons (in no particular order):
1) Many CPUID program writers didn't (and still don't) have a clue how to
calculate the memory speed so did it by random guessing instead of
researching.
2) Sudhian media discovered that ratios and half-integer-multipliers didn't
give the results they would expect from simple calculations, and did a
complete train-wreck of an article. It contradicts itself multiple times and
only serves to confuse people more. At the time, I (among many others) knew
exactly how to calculate the real memory frequency, and pointed out to the
author that his results were exactly in line with what would be expected.
3) There's many people out there who THINK they know how it works, don't
really know how it works, but still say they're right anyhow.
> I also read a lot on the Net forums, searching for a way to
> be sure about that frequency. With my way of calculation ...
Hey, maybe AMD got it wrong too! After all, they only made the damn thing.
Look on page 15 of the "AMD Functional Data Sheet, 939 Pin Package"
document. It gives a list of real memory frequencies for various multipliers
and ratio settings. According to you, the memory speed should not change
when you change the multiplier. But look, the 166MHz speeds bounce all over
the place! Care to explain how, using your method, to get a 157.14MHz memory
clock with a 200MHz FSB, a 11x multiplier, and a 166MHz ratio setting? Note
that the standard method has no such trouble.
Additionally, Oskar Wu (who is probably one of the best motherboard design
engineers in the business) did oscilloscope measurements of memory
frequencies at a range of multipliers and ratios:
http://www.xtremesystems.org/forums/showthread.php?t=41595
Care to explain these numbers too?
[...]
> - people have serious problems calculating memory clock most "known"
> way when Cpu multiplier is a non integer number like 8,5x !!!
No, not at all. There's no difference between calculating for a 8.0x
multiplier and an 8.5x multiplier. The *only* problem is with a 183MHz ratio
setting, where AMD's engineers appeared to make a mistake in their
calculation program resulting in the wrong divider being used.
> This
> could be linked also to upper problem.
> My type of calculation does NOT have this problem!
Your type of calculation does not give the right answer. According to your
calculation, there should be no drop in memory speed shifting to a
half-integer multiplier. Except there most definately is. Oops.
[...]
The K8 memory speed is not independent of the CPU multiplier, hence cannot
be derived from a simple divisor of the HTT speed. QED.
--
Michael Brown
www.emboss.co.nz : OOS/RSI software and more
Add michael@ to emboss.co.nz ---+--- My inbox is always open