wasting the memory speed

On Tue, 13 Apr 2004 23:24:29 +0200, somebody wrote:

On Tue, 13 Apr 2004 09:19:18 GMT, Wes Newell
wrote:

On Tue, 13 Apr 2004 00:35:58 +0200, somebody wrote:

On Sat, 10 Apr 2004 13:01:39 -0700, "Jim" wrote:

Ok, one problem here is that the marketing hype leads the advertising people
to not quite state things correctly. Let's forget the details of that ad,
and talk as accurately as possible.

The CPU FSB is more correctly stated as 200MHz (actual), but since it
employs DDR technology, it's *effective* rate is 400MHz. The "400MHz"
number you see in the ad is actually misleading, the CPU FSB does NOT
physically *run* at 400MHz, it runs at 200MHz (that's what would show up on
a scope!), *but*, because the DDR technology it employs allows data to
travel BOTH on the up and down side of each cycle (per MHz), it's *behaving*
as if it data was traveling on ONE side of the cycle, but at 400MHz! Get
it? It's a marketing gimic, the CPU isn't actually *faster*, it's more
*efficient* (2x in fact) at the same speed of a 200MHz processor (1x) that
does NOT employ DDR technology.

The clock is just that. It's just a clock. It might run at 'just' 200
MHz, but it's _NOT_ the speed of the bus! The speed of the bus is
400MHz, and there's no marketing gimmick about that. That is indeed
the _speed_!

Bus speed is measured by the clock cycles.

What is that "Bus speed"? And why is it "measured by the clock
cycles"?

Because that's been the standard for years and it's the only thing that
makes any sense. Using a data throughput can not define the properties
of the bus.

Who does this? - Certainly not AMD or Intel. That's just the thing here.

Oh yes they do. You should check data sheets instead of advertising. AMD
defines the bus speed in clock cycles. But I really don't care how they
want to define it. They are not a defining agency.

If that "bus speed" is measured in clock cycles, then it's because that
"bus speed" is the bus clock speed. We're running in circles.

The speed is 200MHz, not double
that or 4 times that. That is the data rate. Data rates are measured in
Bps or bps, not MHz.

Frequency of data transfers should be measured in MHz (Hz), just like
any frequency of any event. Clocks do not own the unit MHz. But sure,
data rate is fine to measure in Bps.

No, data transfers don't have frequencies as defined by the meaning of
Hertz. You are right about one thing. Clocks don't own Hertz. It's also
designated for radio frequencies. And that's it. These were the two fields
that were given the Hertz symbol to honor him for his work in
electromagnetic fields. read your history. there's no mention of data
rates. That's why Bps and bps were defined.

Granted, that the bandwidth no longer depends on bus speed, but data
rate. The cpu clock speed does however depend on the bus speed. Every
cpu has a multiplier that sets the cpu clock speed in accordance with
the FSB speed. And now just a few problems that can arise from calling
the data rate the the FSB speed.

You can forget about stupid, and idiot too, - well, at least in the IQ
sense.
But: - Stubborn? - Oh most certainly. I'm not married! So here goes:

The bandwidth does depend on a 'speed'. How could it possibly not? And
the 'effective FSB speed' is not the same thing as data rate. Data rate
is width of bus times 'effective bus speed'.

So what's the real clock speed of a bus running an effective speed of
400Mhz? Be careful now because you can't answer this question with any
certainty. Why? because effective clock speed is relative to, yep, the
reall FSB clock speed. I tend to let people slide that use the term
though, but it's just as bogus as using the data rate as the speed. At
least when it's used, one can tell that it's not real.

(And I could disagree, if I wanted to, what we see are some few problems
arising from calling 'FSB clock speed' 'FSB speed' ;-).)

'Bus speed' is an unfortunate term, since AMD and Intel use it as short
for 'effective bus speed', in the case of FSB, while it's also
established in use as short for 'bus clock speed'. I only used that term
(bus speed) once in my post, and I propose that we henceforth are
explicit about this and use 'effective FSB speed' respectively 'FSB
clock speed', or our argument will indeed be stupid. So ok, I agree,
maybe marketing haven't made us any immediate favor here.

Like I said, although I don't like the term effective speed, I can live
with it, even though it doesn't really provide any real information
without knowing it's efectiveness conpared to what.:-)

I think I primarily used the term 'speed'. Also as in 'speed of the
bus', which you interpret as 'clockspeed' at every turn. Please don't do
that. I meant speed as speed. As does AMD and Intel.
"Speed" is speed. 'Clock speed' is just a term for the frequency of
the clock pulses. It doesn't own 'speed'.

As I said earlier. Standard practices that have been in place for years
state that bus speeds are measured in clock cycles. Data speeds over that
bus are measured in throughput. And if you want to narrow it down to one
line of the bus, then it would be in bps.

Tell an engineer to build you a MB with a 400MHz bus do you think
they'll know that you really only mean 100, or 200 in AMD's case.

- But isn't that pretty much exactly what DEC did? "Give us a 200MHz
bus" and they came up with the Alpha bus at a 100MHz clock? I would
think anyone actually asking for a 'speed'(sic) is concerned with speed
in terms of performance.

The way the data is taken off the bus has nothing to do with the clock
cycle. The clock cycle is constant. The bus didn't change. Only the way
the data is transported across it.

(And I think any engineers building a mobo would have to work from very
detailed specs, not just a loose MHz figure.)

Unfortunately, they do now.:-)

Use that 400MHz to calculate your cpu speed and see how far off you'll
be.

Why should I? If I know enough to calculate cpu 'clock speed', I'd know
I need a multiplier and an external _CLOCK_. Why should I grasp blindly
for any and all MHz figure floating around?

Yes, you know, I know it, but there's millions or billions that don't. And
that's where the problem is using bogus numbers for the FSB speed.

The DDR rates and "effective bus speed" again in MHz, are terms and
concepts established in language and specs. It might not be to your
liking, but they are technically motivated, and insisting on your ways
is not very constructive. The concepts needs explaining, not dismissing.

And if they hadn't used bogus BS, nothing would need explaining except the
data rates. People new to this wouldn't be all confused by ther bogus
numbers and why things don't add up, etc.

To more often strictly employ 'data rate' and 'Bps' is excellent. I
think you're right about that (still there's a possibility for another
mixup with 'data transfer rate' here). But language develops as language
will. There's not terribly much we can do about it, but explain and try
to avoid misunderstandings. Not using 'speed' when you really mean
'clock speed' is also helpful.

I'm tired of explaining it, over, and over, and over again. Ah, something
else to put on my website when i get time.

SNIP

Frequency of pulses on a clock is measured in MHz. Frequency of data
transfers occurring on a bus is also measured in MHz. I'm perfectly
aware of which one of these is the 'clock speed' or 'bus clock speed'.

Hertz was not designated as a measurement for these pulses of data bits,
and using it as such in error.

deleted some stuff which I don't care about.

And as my last comment on this. When I pinned AMD to the wall, they
admitted that their bogus FSB speeds were just that, by saying that it's
the effective clock rate when compared to a non DDR bus.

I'm pretty sure though, that they didn't admit to any "bogus"? But
rather tried to explain what it meant, in terms of relevant properties?

Maybe I've still got a copy of it..... Nope. Must have trashed it last
time I cleaned out my archive. Might find it in google groups somewhere.
I've posted it before. Yep, Here it is.

From - Tue Nov 5 13:52:15 2002
From:
To:
Date: Tue, 05 Nov 2002 10:06:17 -0800
Subject: Front Side Bus (KMM136148V18444L0KM)

Hi Wes,

Thanks for contacting AMD's Technical Service Center. The stated FSB
speed is the effective frequency. The physical frequency is half the
effective frequency - 100, 133, or 166MHz. Most motherboards will state
this physical frequency rather than the effective frequency.

The reason for this is that, in the past, chipsets would transfer data
once per clock cycle. With the introduction of the Athlon's chipset,
they began transferring data twice per clock cycle, effectively doubling
the amount of data transferred even though the physical frequency
remained constant. This is similar to DDR (Double Data Rate) Technology,
which is used in DDR SDRAM. You will also find similar technology used
in other products as well, such as RDRAM.

Hope this helps.
Best Regards,
Jeff Hanaoka
AMD Technical Service Center


Original Message Follows:
------------------------
You refer to the Athlon FSB speeds as 200mhz, 266mhz, and 333mhz. Is
this really the front side bus speed? I can't find a motherboard that
supports front side bus speeds higher than 166mhz. What gives?

--
Abit KT7-Raid (KT133) Tbred B core CPU @2400MHz (24x100FSB)
http://mysite.verizon.net/res0exft/cpu.htm

On Wed, 14 Apr 2004 05:41:25 GMT, Wes Newell
wrote:

Clocks don't own Hertz. It's also
designated for radio frequencies. And that's it. These were the two fields
that were given the Hertz symbol to honor him for his work in
electromagnetic fields. read your history. there's no mention of data
rates. That's why Bps and bps were defined.

A quick check in my Physics handbook, confirms what I thought I knew:

Derived SI Units:

Quantity: frequency
Definition: s^-1
Unit: Hz Herz

There's no note from IUPAP about Hz being reserved for only some
special purposes.
(Which I'm absolutely confident there would have been, if you actually
were right).
Also, SI does not have any other unit for frequency. Sorry, but I take
that to mean that it is in fact an universally employable unit for
frequency. (just consider AC, sound and light, for instance)

You're entirely right of course, that data rate cannot be measured in
Hz, but I've never claimed anything like that. If you thought that,
then I lost you somewhere.

ancra