"AMD jumps amid signs of gaining share on Intel"
http://today.reuters.com/business/ne...&imageid=&cap=

But looks like some of these securities anal...ysts just don't know
what they are writing.
--begin quote--
Cody Acree, an analyst with Stifel Nicolaus, said that Intel's
problem stems from its inability to produce enough computer chips to
meet demand.
"They are capacity constrained so that there's not much ability to
ship more than they've already shipped," Acree said, adding that the
company's strong third quarter probably took some business from the
fourth quarter.
--end quote--
Does it mean that Intel's Q4 capacity suddenly evaporated because it
was overused in Q3?

NNN

On Sat, 10 Dec 2005 04:05:04 GMT, "(E-Mail Removed)"
<(E-Mail Removed)> wrote:

>"AMD jumps amid signs of gaining share on Intel"
>http://today.reuters.com/business/ne...&imageid=&cap=
>
>But looks like some of these securities anal...ysts just don't know
>what they are writing.
>--begin quote--
> Cody Acree, an analyst with Stifel Nicolaus, said that Intel's
>problem stems from its inability to produce enough computer chips to
>meet demand.
>"They are capacity constrained so that there's not much ability to
>ship more than they've already shipped," Acree said, adding that the
>company's strong third quarter probably took some business from the
>fourth quarter.
>--end quote--
>Does it mean that Intel's Q4 capacity suddenly evaporated because it
>was overused in Q3?

They are apparently still short on chipset capacity though, in this
article,
http://www.reed-electronics.com/elec...CA6288399.html there
are two different stories: CFO Bryant says December is "fairly controlled";
a "spokesman" says 2Q06 before they see "relief".

Also in the article, they are "enlisting the assistance of third-party
providers for its lower end chipsets". Dunno what that means: is Intel
out-sourcing fab capacity or are they expecting mbrd mfrs to switch to
alternate chipsets on the low-end? If the latter, they are basically
saying to VIA, SiS ULI, "hey you can have the junk-end of our business for
a few months"... great news huh?

I believe this could also mean that they are ramping up the chipsets with
dual independent bus and FB-DIMM, quite a different beast from the rest,
and maybe even having a bit of trouble with it. DIB & FB-DIMM is Intel's
*big-play* against AMD in the server space and the DIMMs are apparently
running hot as hell: note the fan and the heatpipes on the buffer section
of the DIMM in the lower pic here:
http://www.tecchannel.de/server/hard...7/index19.html


--
Rgds, George Macdonald

George Macdonald wrote:
> They are apparently still short on chipset capacity though, in this
> article,
> http://www.reed-electronics.com/elec...CA6288399.html there
> are two different stories: CFO Bryant says December is "fairly controlled";
> a "spokesman" says 2Q06 before they see "relief".

Apparently the chipset capacity contraint is caused by the fact they may
be producing their NOR on the same production lines. These recent
alliances with Micron on NAND and STMicro on NOR, might have something
to do with moving the flash stuff off of their own assembly lines.

> Also in the article, they are "enlisting the assistance of third-party
> providers for its lower end chipsets". Dunno what that means: is Intel
> out-sourcing fab capacity or are they expecting mbrd mfrs to switch to
> alternate chipsets on the low-end? If the latter, they are basically
> saying to VIA, SiS ULI, "hey you can have the junk-end of our business for
> a few months"... great news huh?

These days, I don't see anything but low-end from Intel. Pentium 4's are
being mated to low-end chipsets to produce $299 computers; almost
pushing Celeron out of the desktop market entirely, since P4's are so
cheap. Even a lot of their laptop sales are being driven by Celeron-M
rather than Pentium-M, which means they'll need even low-end laptop
chipsets.

AMD just drops some of their lower speed-rated Athlon 64's and they are
well out of Intel's range in terms of performance. So Athlon 64's aren't
really competing against Pentium 4, Pentium 4 is competing against Sempron.

> I believe this could also mean that they are ramping up the chipsets with
> dual independent bus and FB-DIMM, quite a different beast from the rest,
> and maybe even having a bit of trouble with it. DIB & FB-DIMM is Intel's
> *big-play* against AMD in the server space and the DIMMs are apparently
> running hot as hell: note the fan and the heatpipes on the buffer section
> of the DIMM in the lower pic here:
> http://www.tecchannel.de/server/hard...7/index19.html

Can't see how it's going to help them, AMD is bringing out Socket F for
servers in a few months. That's going to be able to do DDR2 as well as
DDR3 when it comes out. So it should be able to do FB-DIMM just as easily.

Yousuf Khan

On Sat, 10 Dec 2005 12:56:20 -0500, Yousuf Khan <(E-Mail Removed)> wrote:

>George Macdonald wrote:
>> They are apparently still short on chipset capacity though, in this
>> article,
>> http://www.reed-electronics.com/elec...CA6288399.html there
>> are two different stories: CFO Bryant says December is "fairly controlled";
>> a "spokesman" says 2Q06 before they see "relief".
>
>Apparently the chipset capacity contraint is caused by the fact they may
>be producing their NOR on the same production lines. These recent
>alliances with Micron on NAND and STMicro on NOR, might have something
>to do with moving the flash stuff off of their own assembly lines.

I thought the NAND deal with Micron was a new venture for both of them...
tempted by the current fad for NAND and to satisfy Apple's appetite.

>> Also in the article, they are "enlisting the assistance of third-party
>> providers for its lower end chipsets". Dunno what that means: is Intel
>> out-sourcing fab capacity or are they expecting mbrd mfrs to switch to
>> alternate chipsets on the low-end? If the latter, they are basically
>> saying to VIA, SiS ULI, "hey you can have the junk-end of our business for
>> a few months"... great news huh?
>
>These days, I don't see anything but low-end from Intel. Pentium 4's are
>being mated to low-end chipsets to produce $299 computers; almost
>pushing Celeron out of the desktop market entirely, since P4's are so
>cheap. Even a lot of their laptop sales are being driven by Celeron-M
>rather than Pentium-M, which means they'll need even low-end laptop
>chipsets.
>
>AMD just drops some of their lower speed-rated Athlon 64's and they are
>well out of Intel's range in terms of performance. So Athlon 64's aren't
>really competing against Pentium 4, Pentium 4 is competing against Sempron.

I dunno where you're seeing this - P4 is not that far behind Athlon64 on
the high end... certainly nothing to get complacent about.

>> I believe this could also mean that they are ramping up the chipsets with
>> dual independent bus and FB-DIMM, quite a different beast from the rest,
>> and maybe even having a bit of trouble with it. DIB & FB-DIMM is Intel's
>> *big-play* against AMD in the server space and the DIMMs are apparently
>> running hot as hell: note the fan and the heatpipes on the buffer section
>> of the DIMM in the lower pic here:
>> http://www.tecchannel.de/server/hard...7/index19.html
>
>Can't see how it's going to help them, AMD is bringing out Socket F for
>servers in a few months. That's going to be able to do DDR2 as well as
>DDR3 when it comes out. So it should be able to do FB-DIMM just as easily.

Huh? All three interfaces on a single chip? I thought FB-DIMM was a
completely different interface. I think you're seriously underestimating
Intel's upcoming server/workstation "platform" and AMD had better get their
finger out with 65nm if they want to stay ahead. I have a feeling that
FB-DIMM is going to score big for Intel there and if AMD has to do another
socket for it, things could get messy.

--
Rgds, George Macdonald

George Macdonald wrote:
>>Apparently the chipset capacity contraint is caused by the fact they may
>>be producing their NOR on the same production lines. These recent
>>alliances with Micron on NAND and STMicro on NOR, might have something
>>to do with moving the flash stuff off of their own assembly lines.
>
>
> I thought the NAND deal with Micron was a new venture for both of them...
> tempted by the current fad for NAND and to satisfy Apple's appetite.

No, Micron was already a producer of NAND, but it was well behind the
marketshare leader, Samsung. I assume that the Intel-Micron linkup will
allow Intel to gain access an existing NAND design, without having
develop their own. While Micron will gain access to some pretty big fab
capacity.

>>These days, I don't see anything but low-end from Intel. Pentium 4's are
>>being mated to low-end chipsets to produce $299 computers; almost
>>pushing Celeron out of the desktop market entirely, since P4's are so
>>cheap. Even a lot of their laptop sales are being driven by Celeron-M
>>rather than Pentium-M, which means they'll need even low-end laptop
>>chipsets.
>>
>>AMD just drops some of their lower speed-rated Athlon 64's and they are
>>well out of Intel's range in terms of performance. So Athlon 64's aren't
>>really competing against Pentium 4, Pentium 4 is competing against Sempron.
>
>
> I dunno where you're seeing this - P4 is not that far behind Athlon64 on
> the high end... certainly nothing to get complacent about.

I'm just telling you what I see in my local stores around here. They're
putting P4's into the cheap $299-$750 computers these days. The only
difference between the cheap $750 and the super-cheap $299 computers is
the amount of RAM and hard disk capacity, but not the processors. The
processors are all generally sub-3.0Ghz P4's. They're competing agains
Semprons, at the same price points.

I don't believe that any of the P4's that are above 3.2Ghz are all that
common. But for AMD, the 3200+ and above Athlon 64's are quite common,
and in fact, prevalent. The Athlon 64's desktops are generally in the
$650-$2100 range: the high end of that range would of course be an X2 or FX.

>>Can't see how it's going to help them, AMD is bringing out Socket F for
>>servers in a few months. That's going to be able to do DDR2 as well as
>>DDR3 when it comes out. So it should be able to do FB-DIMM just as easily.
>
>
> Huh? All three interfaces on a single chip? I thought FB-DIMM was a
> completely different interface. I think you're seriously underestimating
> Intel's upcoming server/workstation "platform" and AMD had better get their
> finger out with 65nm if they want to stay ahead. I have a feeling that
> FB-DIMM is going to score big for Intel there and if AMD has to do another
> socket for it, things could get messy.

No, FB-DIMM's claim to fame is that it can remove the interface
differences between different generations of DRAM, so DDR2 and DDR3
FB-DIMMs will be exactly the same. It's the next generation of buffered
RAM. So it should look like just a buffered DDR2 chip.

And no, FB-DIMM won't be a big score for Intel, it reduces performance,
and increases power consumption. So even though AMD is likely to support
FB-DIMM, it's future is not entirely guaranteed.

Yousuf Khan

On Sun, 11 Dec 2005 01:24:06 -0500, Yousuf Khan <(E-Mail Removed)> wrote:

>George Macdonald wrote:
>>>Apparently the chipset capacity contraint is caused by the fact they may
>>>be producing their NOR on the same production lines. These recent
>>>alliances with Micron on NAND and STMicro on NOR, might have something
>>>to do with moving the flash stuff off of their own assembly lines.
>>
>>
>> I thought the NAND deal with Micron was a new venture for both of them...
>> tempted by the current fad for NAND and to satisfy Apple's appetite.
>
>No, Micron was already a producer of NAND, but it was well behind the
>marketshare leader, Samsung. I assume that the Intel-Micron linkup will
>allow Intel to gain access an existing NAND design, without having
>develop their own. While Micron will gain access to some pretty big fab
>capacity.

I don't think so on fabs - they are (initially) to be in Boise, Manassas,
VA and Lehi, UT, all Micron facilities AFAIK. Obviously this has no impact
on chipset production. The NOR deal with ST hasn't given any indication of
production off-loading either - it seems it's mainly a common specification
where they seem to be trading some IP to come up with a common drop-in
part.

>>>These days, I don't see anything but low-end from Intel. Pentium 4's are
>>>being mated to low-end chipsets to produce $299 computers; almost
>>>pushing Celeron out of the desktop market entirely, since P4's are so
>>>cheap. Even a lot of their laptop sales are being driven by Celeron-M
>>>rather than Pentium-M, which means they'll need even low-end laptop
>>>chipsets.
>>>
>>>AMD just drops some of their lower speed-rated Athlon 64's and they are
>>>well out of Intel's range in terms of performance. So Athlon 64's aren't
>>>really competing against Pentium 4, Pentium 4 is competing against Sempron.
>>
>>
>> I dunno where you're seeing this - P4 is not that far behind Athlon64 on
>> the high end... certainly nothing to get complacent about.
>
>I'm just telling you what I see in my local stores around here. They're
>putting P4's into the cheap $299-$750 computers these days. The only
>difference between the cheap $750 and the super-cheap $299 computers is
>the amount of RAM and hard disk capacity, but not the processors. The
>processors are all generally sub-3.0Ghz P4's. They're competing agains
>Semprons, at the same price points.
>
>I don't believe that any of the P4's that are above 3.2Ghz are all that
>common. But for AMD, the 3200+ and above Athlon 64's are quite common,
>and in fact, prevalent. The Athlon 64's desktops are generally in the
>$650-$2100 range: the high end of that range would of course be an X2 or FX.

I never look in local stores for computers - low expectations anyway and
I'm not into tormenting sales-folk.:-) On-line prices of comparable power
CPUs from both is certainly in the same ballpark.

>>>Can't see how it's going to help them, AMD is bringing out Socket F for
>>>servers in a few months. That's going to be able to do DDR2 as well as
>>>DDR3 when it comes out. So it should be able to do FB-DIMM just as easily.
>>
>>
>> Huh? All three interfaces on a single chip? I thought FB-DIMM was a
>> completely different interface. I think you're seriously underestimating
>> Intel's upcoming server/workstation "platform" and AMD had better get their
>> finger out with 65nm if they want to stay ahead. I have a feeling that
>> FB-DIMM is going to score big for Intel there and if AMD has to do another
>> socket for it, things could get messy.
>
>No, FB-DIMM's claim to fame is that it can remove the interface
>differences between different generations of DRAM, so DDR2 and DDR3
>FB-DIMMs will be exactly the same. It's the next generation of buffered
>RAM. So it should look like just a buffered DDR2 chip.

No, it provides a common interface to DDR2 and DDR3 DRAM *chips* on the
module but at the module/chipset level it's a quite different interface
from the current DDR2 channel - AMD would need a different memory
controller, with a different (serial) memory channel speed, signalling and
pin count to do FB-DIMMs. The FB-DIMM "channel" is as different from
current DDR2's as Rambus' DRDRAM was from DDR-SDRAM's; in fact you *could*
say that FB-DIMM is a "clean room" attempt at an alternative DRDRAM
channel. We'll see what Rambus makes of it as an err, infringer.:-)

>And no, FB-DIMM won't be a big score for Intel, it reduces performance,
>and increases power consumption. So even though AMD is likely to support
>FB-DIMM, it's future is not entirely guaranteed.

Intel seems to disagree with you and the results, even counting some
finagling on benchmarks, do not support your contention of lower
performance... a bit of extra latency gets traded for near-lossless
increases of capacity through DIMM count. For *big* memory MPU systems,
Intel may need it more (than AMD) for the moment but it looks like the way
forward in that space. I've no doubt Intel will get some mileage out of
this in holding server customers. I'm sure that Daytripper has some
insights here but probably can't reveal anything.... yet.:-)

--
Rgds, George Macdonald

On Sun, 11 Dec 2005 07:22:43 -0500, George Macdonald
<fammacd=!SPAM^(E-Mail Removed)> wrote:

>
>Intel seems to disagree with you and the results, even counting some
>finagling on benchmarks, do not support your contention of lower
>performance... a bit of extra latency gets traded for . For *big* memory MPU systems,
>Intel may need it more (than AMD) for the moment but it looks like the way
>forward in that space. I've no doubt Intel will get some mileage out of
>this in holding server customers. I'm sure that Daytripper has some
>insights here but probably can't reveal anything.... yet.:-)
>
>--
>Rgds, George Macdonald

Indeed, when the memory is all hanging off the chipset, it's necessary
to have "near-lossless increases of capacity through DIMM count".
When memory controller is integrated, the memory capacity grows
linearly with CPU count. Usually duallie doesn't need as much RAM as
an 8 socket monster. Just as AMD is well enough with older generation
DDR, handily beating DDR2 systems, it would be able to do without
FB-DIMM... at least for a while until FB-DIMM technology is gone
thtough the unavoidable pain of maturing.

NNN

George Macdonald wrote:
>>No, Micron was already a producer of NAND, but it was well behind the
>>marketshare leader, Samsung. I assume that the Intel-Micron linkup will
>>allow Intel to gain access an existing NAND design, without having
>>develop their own. While Micron will gain access to some pretty big fab
>>capacity.
>
>
> I don't think so on fabs - they are (initially) to be in Boise, Manassas,
> VA and Lehi, UT, all Micron facilities AFAIK. Obviously this has no impact
> on chipset production. The NOR deal with ST hasn't given any indication of
> production off-loading either - it seems it's mainly a common specification
> where they seem to be trading some IP to come up with a common drop-in
> part.

I guess for the time being it's mostly being kept in existing Micron
facilities, since they're already producing them. In the future, it's
likely that some of those big 300mm Intel plants will be producing them.

Also it's likely that their joint-venture will allow either or both
companies to spin-off the division, like AMD and Fujitsu just have with
Spansion.

As for STMicro and Intel, yeah, it may be nothing more than an IP trade.


>>I'm just telling you what I see in my local stores around here. They're
>>putting P4's into the cheap $299-$750 computers these days. The only
>>difference between the cheap $750 and the super-cheap $299 computers is
>>the amount of RAM and hard disk capacity, but not the processors. The
>>processors are all generally sub-3.0Ghz P4's. They're competing agains
>>Semprons, at the same price points.
>>
>>I don't believe that any of the P4's that are above 3.2Ghz are all that
>>common. But for AMD, the 3200+ and above Athlon 64's are quite common,
>>and in fact, prevalent. The Athlon 64's desktops are generally in the
>>$650-$2100 range: the high end of that range would of course be an X2 or FX.
>
>
> I never look in local stores for computers - low expectations anyway and
> I'm not into tormenting sales-folk.:-) On-line prices of comparable power
> CPUs from both is certainly in the same ballpark.

But the Intel platforms are much more expensive, even if the processors
themselves are comparable. That's because of the confusing "new chipset
for every new chip" situation, and the need to add additional cooling to
Intel platforms.

I've not seen a lot of the fastest Intel P4's in any of the retail
spaces. Not in big-box stores like Future Shop, and not in little
mom'n'pops either.

>>>>Can't see how it's going to help them, AMD is bringing out Socket F for
>>>>servers in a few months. That's going to be able to do DDR2 as well as
>>>>DDR3 when it comes out. So it should be able to do FB-DIMM just as easily.
>>>
>>>
>>>Huh? All three interfaces on a single chip? I thought FB-DIMM was a
>>>completely different interface. I think you're seriously underestimating
>>>Intel's upcoming server/workstation "platform" and AMD had better get their
>>>finger out with 65nm if they want to stay ahead. I have a feeling that
>>>FB-DIMM is going to score big for Intel there and if AMD has to do another
>>>socket for it, things could get messy.
>>
>>No, FB-DIMM's claim to fame is that it can remove the interface
>>differences between different generations of DRAM, so DDR2 and DDR3
>>FB-DIMMs will be exactly the same. It's the next generation of buffered
>>RAM. So it should look like just a buffered DDR2 chip.
>
>
> No, it provides a common interface to DDR2 and DDR3 DRAM *chips* on the
> module but at the module/chipset level it's a quite different interface
> from the current DDR2 channel - AMD would need a different memory
> controller, with a different (serial) memory channel speed, signalling and
> pin count to do FB-DIMMs. The FB-DIMM "channel" is as different from
> current DDR2's as Rambus' DRDRAM was from DDR-SDRAM's; in fact you *could*
> say that FB-DIMM is a "clean room" attempt at an alternative DRDRAM
> channel. We'll see what Rambus makes of it as an err, infringer.:-)

The pinouts from FB-DIMM might be different, just like registered and
non-registered DDR is different. But all of the same signals are there,
and it's just a matter of the motherboard makers laying out the traces
properly in each case.

It took AMD just one additional pin to support DDR2 from DDR in its
desktop line when it goes from Socket 939 to Socket M2 (940-pin).
However, in its server line, it's going from Socket 940 to Socket F
(1206-pin). There's a lot more pins there so some of it must be reserved
for future ram like DDR3 and FB-DIMM.

>>And no, FB-DIMM won't be a big score for Intel, it reduces performance,
>>and increases power consumption. So even though AMD is likely to support
>>FB-DIMM, it's future is not entirely guaranteed.
>
>
> Intel seems to disagree with you and the results, even counting some
> finagling on benchmarks, do not support your contention of lower
> performance... a bit of extra latency gets traded for near-lossless
> increases of capacity through DIMM count. For *big* memory MPU systems,
> Intel may need it more (than AMD) for the moment but it looks like the way
> forward in that space. I've no doubt Intel will get some mileage out of
> this in holding server customers. I'm sure that Daytripper has some
> insights here but probably can't reveal anything.... yet.:-)

Higher latency, and the buffer itself adds to the wattage consumed. I
think I read it's expected to as 10W per DIMM!

Yousuf Khan

On Sun, 11 Dec 2005 15:46:18 GMT, "(E-Mail Removed)"
<(E-Mail Removed)> wrote:

>On Sun, 11 Dec 2005 07:22:43 -0500, George Macdonald
><fammacd=!SPAM^(E-Mail Removed)> wrote:
>
>>
>>Intel seems to disagree with you and the results, even counting some
>>finagling on benchmarks, do not support your contention of lower
>>performance... a bit of extra latency gets traded for . For *big* memory MPU systems,
>>Intel may need it more (than AMD) for the moment but it looks like the way
>>forward in that space. I've no doubt Intel will get some mileage out of
>>this in holding server customers. I'm sure that Daytripper has some
>>insights here but probably can't reveal anything.... yet.:-)
>>
>>--
>>Rgds, George Macdonald
>
>Indeed, when the memory is all hanging off the chipset, it's necessary
>to have "near-lossless increases of capacity through DIMM count".
>When memory controller is integrated, the memory capacity grows
>linearly with CPU count.

Obviously that's what I meant by Intel possibly needing it more than AMD...
*BUT* if you look at Intel's chipsets in that space, they have multiple
channels "hanging off the chipset" anyway, so up to a certain memory size
it's six and half a dozen with a CPU socket count of 2 or 4. Above some
memory load, you just have to slow the memory channel clocks -- common with
AMD CPUs with >4 ranks of memory -- and there is an absolute ceiling.

> Usually duallie doesn't need as much RAM as
>an 8 socket monster. Just as AMD is well enough with older generation
>DDR, handily beating DDR2 systems, it would be able to do without
>FB-DIMM... at least for a while until FB-DIMM technology is gone
>thtough the unavoidable pain of maturing.

It's going to happen. Besides the advantages of FB-DIMM, such as
simultaneous read/write, the size issue is on the horizon already,
especially when you consider the slowing required to get a full DIMM count
on an Opteron. In fact at that point DDR does not beat DDR2 "handily".
Apart from practical issues, this is just another item for the anal...ysts
to latch on to in boosting Intel. AMD needs it on their roadmap.

--
Rgds, George Macdonald

On Sun, 11 Dec 2005 14:35:55 -0500, Yousuf Khan <(E-Mail Removed)> wrote:

>George Macdonald wrote:
>>>No, Micron was already a producer of NAND, but it was well behind the
>>>marketshare leader, Samsung. I assume that the Intel-Micron linkup will
>>>allow Intel to gain access an existing NAND design, without having
>>>develop their own. While Micron will gain access to some pretty big fab
>>>capacity.
>>
>>
>> I don't think so on fabs - they are (initially) to be in Boise, Manassas,
>> VA and Lehi, UT, all Micron facilities AFAIK. Obviously this has no impact
>> on chipset production. The NOR deal with ST hasn't given any indication of
>> production off-loading either - it seems it's mainly a common specification
>> where they seem to be trading some IP to come up with a common drop-in
>> part.
>
>I guess for the time being it's mostly being kept in existing Micron
>facilities, since they're already producing them. In the future, it's
>likely that some of those big 300mm Intel plants will be producing them.

Flash is a highly cyclical business so far - speculation is just that.

>Also it's likely that their joint-venture will allow either or both
>companies to spin-off the division, like AMD and Fujitsu just have with
>Spansion.

You mean like an IPO of IM Flash Technologies LLC?:-)

>As for STMicro and Intel, yeah, it may be nothing more than an IP trade.
>
>
>>>I'm just telling you what I see in my local stores around here. They're
>>>putting P4's into the cheap $299-$750 computers these days. The only
>>>difference between the cheap $750 and the super-cheap $299 computers is
>>>the amount of RAM and hard disk capacity, but not the processors. The
>>>processors are all generally sub-3.0Ghz P4's. They're competing agains
>>>Semprons, at the same price points.
>>>
>>>I don't believe that any of the P4's that are above 3.2Ghz are all that
>>>common. But for AMD, the 3200+ and above Athlon 64's are quite common,
>>>and in fact, prevalent. The Athlon 64's desktops are generally in the
>>>$650-$2100 range: the high end of that range would of course be an X2 or FX.
>>
>>
>> I never look in local stores for computers - low expectations anyway and
>> I'm not into tormenting sales-folk.:-) On-line prices of comparable power
>> CPUs from both is certainly in the same ballpark.
>
>But the Intel platforms are much more expensive, even if the processors
>themselves are comparable. That's because of the confusing "new chipset
>for every new chip" situation, and the need to add additional cooling to
>Intel platforms.
>
>I've not seen a lot of the fastest Intel P4's in any of the retail
>spaces. Not in big-box stores like Future Shop, and not in little
>mom'n'pops either.

When I swing by CompUSA or BestBuy to pick up ink cartridges etc. and I see
the prospective buyers in retail I'm not surprised by that. On-line dwarfs
retail of course anyway and at the consumer high end, gamers are hot on AMD
just now; in corporate high end seems to be where P4 has its mainstay. In
fact consumer high-end, largely gamers, is hot on DIY boxes anyway.

>>>>>Can't see how it's going to help them, AMD is bringing out Socket F for
>>>>>servers in a few months. That's going to be able to do DDR2 as well as
>>>>>DDR3 when it comes out. So it should be able to do FB-DIMM just as easily.
>>>>
>>>>
>>>>Huh? All three interfaces on a single chip? I thought FB-DIMM was a
>>>>completely different interface. I think you're seriously underestimating
>>>>Intel's upcoming server/workstation "platform" and AMD had better get their
>>>>finger out with 65nm if they want to stay ahead. I have a feeling that
>>>>FB-DIMM is going to score big for Intel there and if AMD has to do another
>>>>socket for it, things could get messy.
>>>
>>>No, FB-DIMM's claim to fame is that it can remove the interface
>>>differences between different generations of DRAM, so DDR2 and DDR3
>>>FB-DIMMs will be exactly the same. It's the next generation of buffered
>>>RAM. So it should look like just a buffered DDR2 chip.
>>
>>
>> No, it provides a common interface to DDR2 and DDR3 DRAM *chips* on the
>> module but at the module/chipset level it's a quite different interface
>> from the current DDR2 channel - AMD would need a different memory
>> controller, with a different (serial) memory channel speed, signalling and
>> pin count to do FB-DIMMs. The FB-DIMM "channel" is as different from
>> current DDR2's as Rambus' DRDRAM was from DDR-SDRAM's; in fact you *could*
>> say that FB-DIMM is a "clean room" attempt at an alternative DRDRAM
>> channel. We'll see what Rambus makes of it as an err, infringer.:-)
>
>The pinouts from FB-DIMM might be different, just like registered and
>non-registered DDR is different. But all of the same signals are there,
>and it's just a matter of the motherboard makers laying out the traces
>properly in each case.

No, it still has 240-pin form factor but it's not "just like......" and it
doesn't have "the same signals"... see below:

>It took AMD just one additional pin to support DDR2 from DDR in its
>desktop line when it goes from Socket 939 to Socket M2 (940-pin).
>However, in its server line, it's going from Socket 940 to Socket F
>(1206-pin). There's a lot more pins there so some of it must be reserved
>for future ram like DDR3 and FB-DIMM.

I suggest you take a look at some datasheets - the signalling for FB-DIMM
is, as I already said, completely different: there's northbound and
southbound serial links with 14 and 10 lanes respectively. It's a lot more
than a variation of registering and there's no way that AMD is going to be
able to, or even want to, build in DDR2 and FB-DIMM support into a single
chip.

>>>And no, FB-DIMM won't be a big score for Intel, it reduces performance,
>>>and increases power consumption. So even though AMD is likely to support
>>>FB-DIMM, it's future is not entirely guaranteed.
>>
>>
>> Intel seems to disagree with you and the results, even counting some
>> finagling on benchmarks, do not support your contention of lower
>> performance... a bit of extra latency gets traded for near-lossless
>> increases of capacity through DIMM count. For *big* memory MPU systems,
>> Intel may need it more (than AMD) for the moment but it looks like the way
>> forward in that space. I've no doubt Intel will get some mileage out of
>> this in holding server customers. I'm sure that Daytripper has some
>> insights here but probably can't reveal anything.... yet.:-)
>
>Higher latency, and the buffer itself adds to the wattage consumed. I
>think I read it's expected to as 10W per DIMM!

Yeah well the heatpipes on the AMB sections are not a good sign but those
are prototypes or samples so it's early days yet. As for latency, there's
wins & losses... simultaneous read/write could count for a lot in a server
system.

--
Rgds, George Macdonald