Latest Athlon 64 product introductions

[spinlock]

You know what they say,

"two Eds are better than one"

 

[KR Williams]

939 boards are also cheaper to make, 4 layer instead of 6.

Why? I don't see the memory differences as significant.

 

[KR Williams]

@twister01.bloor.is.net.cable.rogers.com>, (e-mail address removed)
says...

Oh geez, now we got two Ed's, and they're gonna carry on a confusing debate
amongst themselves.

Nope. Same Ed, just a different moniker (look at the headings).

 

[Yousuf Khan]

@twister01.bloor.is.net.cable.rogers.com>, (e-mail address removed)
says...

Nope. Same Ed, just a different moniker (look at the headings).

If they are the same Ed, then why is one asking a question of the other?
This has the potential of making the Tony vs. Tony Hill debates look
comprehensible.

Yousuf Khan

 

[Rupert Pigott]

Why? I don't see the memory differences as significant.

Same question crossed my mind. My guess is that they weren't *too*
bothered about Opteron boards being costly so they didn't put much
effort into keeping the layer count down. I'll bet that some bright
spark spotted an alternative pin-out that would allow 4 layer
boards sometime after S940 was released, hence the magic new S939
that allows 4-Layer boards.

Cheers,
Rupert

 

[Rob Stow]

Same question crossed my mind. My guess is that they weren't *too*
bothered about Opteron boards being costly so they didn't put much
effort into keeping the layer count down. I'll bet that some bright
spark spotted an alternative pin-out that would allow 4 layer
boards sometime after S940 was released, hence the magic new S939
that allows 4-Layer boards.

No. Socket 940, 939, and 754 were all announced at the same time.
Manufacturers were ready to start cranking out Socket 939 boards
more than a year ago but AMD just wasn't producing the chips.

As well, PDF illustrating the pin-outs for all three sockets were
available at AMD's site early last year and I would expect they
are still there if you are interested.

 

[Rupert Pigott]

No. Socket 940, 939, and 754 were all announced at the same time.
Manufacturers were ready to start cranking out Socket 939 boards
more than a year ago but AMD just wasn't producing the chips.

Eeek, I hadn't twigged 939 & 940 were launched at the same time. Seems
a bit odd that the 4/6 layer difference exists in that case.

As well, PDF illustrating the pin-outs for all three sockets were
available at AMD's site early last year and I would expect they
are still there if you are interested.

Must admit I'm curious to see what kind of differences exist between 939
and 940.

Thanks,
Rupert

 

[GSV Three Minds in a Can]

Bitstring <[email protected]>, from the wonderful

Same question crossed my mind. My guess is that they weren't *too*
bothered about Opteron boards being costly so they didn't put much
effort into keeping the layer count down. I'll bet that some bright
spark spotted an alternative pin-out that would allow 4 layer
boards sometime after S940 was released, hence the magic new S939
that allows 4-Layer boards.

I'm not sure you can blame the pinout - I can't see why you can't route
anything with 4 layers (actually, iirc, with 2 .. in theory, for
constant width wires), unless/until you start needing massive power
distribution and ground planes?

 

[Yousuf Khan]

Must admit I'm curious to see what kind of differences exist between
939 and 940.

From a purely end-user perspective, the biggest difference between 939 and
940 is that 939 is for unbuffered DDR, while 940 is for server-class
buffered DDR. I assume that since 940 uses buffered DDR, the potential
exists to outfit 940 board with well over two DIMMs per processor.

Yousuf Khan

 

[GSV Three Minds in a Can]

Read that last sentence again. ;-)

The reality is that you need the ground/power planes for electrical
reasons, other than the *MASSIVE* power these things dissipate.
Think about 60W at ~1.25V. That's a wee bit of current. In
addition the impedance of the power distribution must be kept as
low as possible to reduce noise. The planes are also necessary to
keep the impedance of the signal lines constant. Two of the four
planes were power planes in the 486 days, without the massive
currents we now see.

Yes, you can theoretically wire anything using two layers, with an
infinite wiring space, infinitely long wires, and an infinite
number of vias. To make everything work, there are restrictions on
all of these as well as the differences in these from one wire to
another.

Yes I know .. I spent 20 years working on / managing IC CAD software,
including routers, and I haven't quiet forgotten all of it yet. 8>.

You still haven't explained to my satisfaction why a =pin-out change=
can suddenly force the requirement for another two wiring planes. I can
see why an =additional= power requirement, or need for =additional=
signal conditioning (like you're running on the hairy edge, which is why
you needed buffered DIMMS in the first place) could up the wiring planes
needed, but a pinout change which magically needs two extra board layers
seems like something you could hardly devise if you worked at it.

 

[K Williams]

If they are the same Ed, then why is one asking a question of the
other?

My news server has a "Grumble" post threaded in between.

This has the potential of making the Tony vs. Tony Hill
debates look comprehensible.

;-) That one got me!

 

[K Williams]

Same question crossed my mind. My guess is that they weren't *too*
bothered about Opteron boards being costly so they didn't put much
effort into keeping the layer count down. I'll bet that some
bright spark spotted an alternative pin-out that would allow 4
layer boards sometime after S940 was released, hence the magic new
S939 that allows 4-Layer boards.

The only possible difference I can see is that the S939 boards won't
be pretending to do more than one processor, so may be somewhat
simpler that way. However that would assume the HT links aren't
being used for anything else. So why all the pins?

 

[K Williams]

From a purely end-user perspective, the biggest difference between
939 and 940 is that 939 is for unbuffered DDR, while 940 is for
server-class buffered DDR. I assume that since 940 uses buffered
DDR, the potential exists to outfit 940 board with well over two
DIMMs per processor.

But, but... Why would a 940 board that *only* supports four DIMMs
be more complicated than a 939 board that only supports 4 DIMMs?
DIMMs is Dimms (at least on this level).

 

[K Williams]

Bitstring <[email protected]>, from the
wonderful person Rupert Pigott


I'm not sure you can blame the pinout - I can't see why you can't
route anything with 4 layers (actually, iirc, with 2 .. in theory,
for constant width wires), unless/until you start needing massive
power distribution and ground planes?

Read that last sentence again. ;-)

The reality is that you need the ground/power planes for electrical
reasons, other than the *MASSIVE* power these things dissipate.
Think about 60W at ~1.25V. That's a wee bit of current. In
addition the impedance of the power distribution must be kept as
low as possible to reduce noise. The planes are also necessary to
keep the impedance of the signal lines constant. Two of the four
planes were power planes in the 486 days, without the massive
currents we now see. Thus, a four-layer board has only two wiring
planes, which are normally wired horizontally and vertically.
"Manhattan" wiring (streets on one side, avenues on the other) to
hook everything together.

Yes, you can theoretically wire anything using two layers, with an
infinite wiring space, infinitely long wires, and an infinite
number of vias. To make everything work, there are restrictions on
all of these as well as the differences in these from one wire to
another.

 

[Rob Stow]

Eeek, I hadn't twigged 939 & 940 were launched at the same time. Seems
a bit odd that the 4/6 layer difference exists in that case.

I saw my first Socket 939 board at the same time I saw my
first 940 board - in Feb 1993. At that time the AMD rep
at the demo was predicting general availability of socket
940 chips for April 2003 (which actually happened) and
socket 939 chips for Q3 2003. So far Q3 2004 is just a
few weeks away and no socket 939 chips seem to be in stock
in any stores.

(I also saw a 4-way board by NewiSys at that demo with
four 800 MHz Opties on it. )

 

[Rob Stow]

Yousuf Khan wrote:




But, but... Why would a 940 board that *only* supports four DIMMs
be more complicated than a 939 board that only supports 4 DIMMs?
DIMMs is Dimms (at least on this level).

There is nothing inherent in the Opty that limits motherboard
manufacturers to 4 DIMMs per processor. HP, for example, has
managed 8 in their 4-way systems.

 

[K Williams]

Yes I know .. I spent 20 years working on / managing IC CAD
software, including routers, and I haven't quiet forgotten all of
it yet. 8>.

You still haven't explained to my satisfaction why a =pin-out
change= can suddenly force the requirement for another two wiring
planes.

Oh, my... We're now asking the same question! My proposal is that
there is no pretense of a MP system with S939, so things are
relaxed a bit. That supposes that a Uni-S940 should be do-able
w/4-layers too. Maybe no one wants to design a S940 board for a
uni? I find that hard to swallow too. so...

I can see why an =additional= power requirement, or need
for =additional= signal conditioning (like you're running on the
hairy edge, which is why you needed buffered DIMMS in the first
place) could up the wiring planes needed, but a pinout change
which magically needs two extra board layers seems like something
you could hardly devise if you worked at it.

Registered DIMMs should need *less* care in layout, given the same
number. That's sorta the purpose behind registering.

 

[Yousuf Khan]

But, but... Why would a 940 board that *only* supports four DIMMs
be more complicated than a 939 board that only supports 4 DIMMs?
DIMMs is Dimms (at least on this level).

Probably because a 940 board will truly be able to support 4 DIMMs whereas a
939 board will more than likely be limited to two DIMMs in reality. When was
the last time you saw desktop boards routinely support more than two DIMMs?
I can only remember this happening back in the Pentium 1 days. Ever since
then, all I've ever seen is warnings against putting any more than two DIMMs
on any board even if it has space for more.

The only ones that can truly do it these days are the buffered DIMMs on
server boards.

Yousuf Khan

 

[Yousuf Khan]

My news server has a "Grumble" post threaded in between.

Yes, so does mine.

But the first Ed was saying that a Socket 745 required six layer mobos,
while Socket 939 requires only 4 layer. Then Grumble asked Ed #1 why that is
the case? Then Ed #2 responded to Grumble by answering with a
semi-rhetorical question: he asked if it wasn't true that S745 actually
requires 4 layers while S939 requires 6 layers, therefore contradicting Ed
#1.

;-) That one got me!

I think if this sort of thing happens more often, then people who are using
things like OE-Quotefix to format their news reader styles, make sure that
they include at least the email addresses in the quote header.

Yousuf Khan

 

[Yousuf Khan]

The only possible difference I can see is that the S939 boards won't
be pretending to do more than one processor, so may be somewhat
simpler that way. However that would assume the HT links aren't
being used for anything else. So why all the pins?

Well as some people have pointed out in the case of Pentium 4's Socket
775 -- extra grounding.

Then again it's possible that maybe AMD is already preparing for DDR2, and
Socket 939 is a forward looking design preparing for that day.

Yousuf Khan