Intel backs away from FB-DIMMs

[YKhan]

Intel pulls back from FB-DIMMs
"Intel is telling everyone who will listen that they are going back to
rDIMMs in 2008 over about 90% of their server line. This is due to them
not being able to hit power numbers, seemingly a problem with the tech
itself vs implementations, and of course cost."
http://www.theinquirer.net/default.aspx?article=34220

Yousuf Khan

 

[nobody]

Intel pulls back from FB-DIMMs
"Intel is telling everyone who will listen that they are going back to
rDIMMs in 2008 over about 90% of their server line. This is due to them
not being able to hit power numbers, seemingly a problem with the tech
itself vs implementations, and of course cost."
http://www.theinquirer.net/default.aspx?article=34220

Yousuf Khan

And one more reason: unified memory platform for server/desktop would
allow Intel to repackage some Xeons as Extremes if need be. As things
stand now, Intel has no answer to 4x4 because the only dual-capable
Intel chipset works only with FB-DIMMs - not optimal for desktop,
especially gaming, aside from high heat. For more details:
http://theinquirer.net/default.aspx?article=33269

NNN

 

[George Macdonald]

Intel pulls back from FB-DIMMs
"Intel is telling everyone who will listen that they are going back to
rDIMMs in 2008 over about 90% of their server line. This is due to them
not being able to hit power numbers, seemingly a problem with the tech
itself vs implementations, and of course cost."
http://www.theinquirer.net/default.aspx?article=34220

OMG does this mean that Intel will pull all those $$ back out of Micron...
AGAIN?

 

[Yousuf Khan]

OMG does this mean that Intel will pull all those $$ back out of Micron...
AGAIN?

No, it just means it's going to punish Micron by having it buyout its
NOR flash division.

Yousuf Khan

 

[David Kanter]

And one more reason: unified memory platform for server/desktop would

allow Intel to repackage some Xeons as Extremes if need be. As things
stand now, Intel has no answer to 4x4

They don't need an answer. 4x4 is a gimmick.

DK

 

[nobody]

They don't need an answer. 4x4 is a gimmick.

DK

Don't be judgemental. When it comes out and gets benchmarked we'll
see. Your opinion might or might not turn out right. So far I see
this upcoming technology as a good alternative to dual Opteron for
high end desktop or midrange workstation, with opty ruling high end.

NNN

 

[YKhan]

They don't need an answer. 4x4 is a gimmick.

Rahul Sood seems to think this might have potential for gamers.

"With the help of Asus, Nvidia, and others, AMD is creating an entirely
new motherboard concept for 4x4. It's not your standard-issue
workstation motherboard; rather, it's an enthusiast-class
multiprocessor motherboard with support for unbuffered non-ECC
enthusiast-class memory. AMD also promises to release three tiers of
4x4 in the box where processor kits will start "well under $1,000"
and go up from there."
http://voodoopc.blogspot.com/2006/09/amd-creating-new-category-of.html

Yousuf Khan

 

[Rick Jones]

workstation motherboard; rather, it's an enthusiast-class
multiprocessor motherboard with support for unbuffered non-ECC
enthusiast-class memory. AMD also promises to release three tiers of

non-ECC memory?!?

rick jones

 

[nobody]

Rahul Sood seems to think this might have potential for gamers.

"With the help of Asus, Nvidia, and others, AMD is creating an entirely
new motherboard concept for 4x4. It's not your standard-issue
workstation motherboard; rather, it's an enthusiast-class
multiprocessor motherboard with support for unbuffered non-ECC
enthusiast-class memory. AMD also promises to release three tiers of
4x4 in the box where processor kits will start "well under $1,000"
and go up from there."
http://voodoopc.blogspot.com/2006/09/amd-creating-new-category-of.html

Yousuf Khan

Does this mean that only FX-class chips have that 2nd cHT link? Then
it would be, at least for me, a tough choice between low-end Opty and
FX - whichever would have better performance for *less*. I hoped that
some mobo makers would sell a stripped-down 4x4 board, possibly with
only 1 socket connected to RAM, on the cheap. 2x Sempron on such a
board would give performance similar to a single dual core A64 for
less, and the upgrade path is good - to A64 and eventually to K8L when
they respectively drop in price. We'll see if cheap A64/Semprons have
that cHT and can work in dual config.
NNN

 

[YKhan]

non-ECC memory?!?

Just the regular type of RAM modules we all use in PC environments. No
error correction capabilities.

Yousuf Khan

 

[YKhan]

Does this mean that only FX-class chips have that 2nd cHT link? Then
it would be, at least for me, a tough choice between low-end Opty and
FX - whichever would have better performance for *less*. I hoped that
some mobo makers would sell a stripped-down 4x4 board, possibly with
only 1 socket connected to RAM, on the cheap. 2x Sempron on such a
board would give performance similar to a single dual core A64 for
less, and the upgrade path is good - to A64 and eventually to K8L when
they respectively drop in price. We'll see if cheap A64/Semprons have
that cHT and can work in dual config.

Where in this article did you get the impression it's only for FX? As
for attaching only one processor to RAM, that's entirely upto you, you
can do that, and the other processor can get its data through
Hypertransport, but you'll be crippling one of the processors.

AMD has already admitted that X2 is also an option for 4x4, so perhaps
that's where the cheapest 4x4 platforms will come from It's also
possible that single-core A64 or Sempron will work in this setup, but
then it would be cheaper to simply go with a single dual-core A64. I
don't agree with you that it would be cheaper to go with two
single-core Semprons vs. one low-end X2. The cheapest X2 is now below
$100, the 3600+.

Yousuf Khan

 

[David Kanter]

They don't need an answer. 4x4 is a gimmick.

Don't be judgemental.

When it comes out and gets benchmarked we'll
see.

Some of us actually look beyond benchmarks. Unlike the AthlonFX or
Core2Duo EE, this requires an entirely new motherboard, that will only
sell in low volume. Having a beefed up CPU is a lot easier, since you
can probably recycle a server CPU, and maybe even one with a fault in
the cache.

By going with a new system they need to:

1. Design a new motherboard
2. Figure out positioning and pricing such that they do not screw up
their existing Opteron product lines

That's a lot tougher than just adding a CPU to the SKU mix. I don't
see any compelling workloads for the system, and it all sounds like
it's more effort than it's worth. If you don't see the marketing
problems this causes, then you really need to think about the problem
more.

Your opinion might or might not turn out right. So far I see
this upcoming technology as a good alternative to dual Opteron for
high end desktop or midrange workstation, with opty ruling high end.

That's the problem.

DK

 

[Yousuf Khan]

By going with a new system they need to:

1. Design a new motherboard
2. Figure out positioning and pricing such that they do not screw up
their existing Opteron product lines

The Opteron 100-series is already Socket AM2-based, so this might be an
option for some workstation classes, or even some cheapie servers
derived off of desktop components like the kinds that IBM and Dell have
been selling for years with Celeron CPUs in them.

Yousuf Khan

 

[nobody]

As
for attaching only one processor to RAM, that's entirely upto you, you
can do that, and the other processor can get its data through
Hypertransport, but you'll be crippling one of the processors.

This might make the motherboard a lot cheaper, and also smaller
footprint. As of now, most dual Opty boards with both sockets
connected to RAM banks are EATX format, while the ones with only one
socket connected are plain ATX. And it's not exactly "crippling" - a
couple of years ago I saw MSI Master2-far (only one socket conneted)
benchmarked against Tyan Thunder K8W (both sockets connected), and it
was essentially a toss-up, with each benchmark results being different
only a few percentage points, and not always in favor of K8W. The
price difference was $200-- vs. $400++, and MSI fit into ATX case,
whereas Tyan required EATX (another $100 or so). Don't forget extra
RAM you'd have to buy to take advantage of both sockets connected.
That influenced my decision to go with MSI, which in the end turned
out a mistake - Tyan is dual core capable, and MSI is not, but who
could've known in early 2004?

NNN

 

[David Kanter]

The Opteron 100-series is already Socket AM2-based, so this might be an
option for some workstation classes, or even some cheapie servers
derived off of desktop components like the kinds that IBM and Dell have
been selling for years with Celeron CPUs in them.

Try thinking about this from a marketing perspective, and you'll
understand why it's a bad idea. If web hosting companies start buying
low-end CPUs for these dual socket systems, AMD won't be making much
money at all, where they did before.

The world doesn't revolve around technology; if you cannot make money
off it, it's not very interesting.

DK

 

[YKhan]

Try thinking about this from a marketing perspective, and you'll
understand why it's a bad idea. If web hosting companies start buying
low-end CPUs for these dual socket systems, AMD won't be making much
money at all, where they did before.

The world doesn't revolve around technology; if you cannot make money
off it, it's not very interesting.

I am thinking of it from the marketing perspective, and it's a
wonderful idea. You're the one who isn't seeing the marketing side of
the equation. IBM and Dell and others have already been selling
cheapskate servers which are simply desktop systems rebadged as
servers, they come with desktop CPUs (e.g. Pentium or even Celeron),
rather than Xeon.

In fact, Intel has even acknowledged this market by creating the new
Xeon 3100-series, which is just a rebadged desktop Core 2 Duo chip.
Meaning that it's a Conroe rather than a Woodcrest (Xeon 5100-series)
or Bensley (Xeon 5000-series). AMD will stake its claim into this same
market by selling rebadged Athlon 64's as the Opteron 1000-series. It's
all part of the two companies' counterpunching, they will not leave any
market alone that the other one has a presense in.

Besides, this comes from the very foundations of the whole x86 server
industry. The whole industry got started back in the 80's, because
people were turning desktops into servers (often by simply replacing
DOS with Netware at the time). It's quite obvious that there is still a
market for such simple no-frills servers. With both Intel and AMD
rebadging their desktop chips as server chips, you get to use cheap
desktop boards, with cheap desktop memory, and cheap desktop hard
disks, etc. The advantage for a customer might be that AMD and Intel
will give these processors some additional server-level warranty
support. The advantage to Intel and AMD would be that they get to
charge a little bit more for the additional support (but not too much
more).

With 4x4 motherboards out there, this might give AMD an opportunity to
open up a market not served by Intel yet -- the no-frills dual-socket
server market. Again this will be all desktop parts. Slightly up from
the no-frills single-socket market. It'll also be different than the
professional dual-socket market, which will use Opteron 2000-series
processors (Socket F), rather than Opteron 1000 (Socket AM2). Opteron
2000 requires all of that expensive registered DIMMs crap, Opteron 1000
won't. Registered memory is useful when you need a *LOT* of memory,
but not so much if you don't.

The secret to riding out the ups'n'downs of the cyclical market is to
have your toes dipped into every market. They're not all going to be
down at the same time.

Yousuf Khan

 

[nobody]

Try thinking about this from a marketing perspective, and you'll
understand why it's a bad idea. If web hosting companies start buying
low-end CPUs for these dual socket systems, AMD won't be making much
money at all, where they did before.

The world doesn't revolve around technology; if you cannot make money
off it, it's not very interesting.

<snip/>

If buying decisions were made on performance only, Conroe would take
the 2p market away from Opteron. However price is a big part of the
picture. If one could live without Registered memory (web servers
sure coud, ditto low and midrange workstations and some DB servers too
- such as development servers) 4x4 would offer savings too significant
to be ignored. Each sale taken from Xeon is a win for AMD, even if
it's for lower margin A64 or even (cough) Sempron. While this would
surely cannibalize some 2p Opteron as well, it is a bigger threat to
traditionally pricier Xeon that goes with yet pricier FB-DIMMs. Intel
would have to either consede market share or drastically lower Xeon
and accompanying chipset price - both ways good for the competitor.
As it turned out, Intel cash hoard has a finite size - otherwise why
would Intel cut 10000+ heads hoping to save $3BN by 2008 - a puny
number for Intel just a few years ago. If AMD takes the price war to
2p segment, it would lighten Intel's competitive pressure in other
(desktop, mobile) segments.

4x4 would also offer some (not much but still) performance advantage
over Opteron because consumer parts were most of the time a speed
grade or two ahead of server parts, and also because unbuffered memory
is faster than buffered. If Opteron slightly trails Xeon in 2p, 4x4
may get closer in performance to Xeon, or even get ahead.

Again, I suggest to reserve the final judgement on this platform, as
on any other, until after it is released. Until then any opinion,
including mine, is nothing more than speculation.

NNN

 

[David Kanter]

I am thinking of it from the marketing perspective,

Not very hard, apparently.

and it's a
wonderful idea. You're the one who isn't seeing the marketing side of
the equation. IBM and Dell and others have already been selling
cheapskate servers which are simply desktop systems rebadged as
servers, they come with desktop CPUs (e.g. Pentium or even Celeron),
rather than Xeon.

Notice how these are 1S systems.

In fact, Intel has even acknowledged this market by creating the new
Xeon 3100-series, which is just a rebadged desktop Core 2 Duo chip.
Meaning that it's a Conroe rather than a Woodcrest (Xeon 5100-series)
or Bensley (Xeon 5000-series).

You mean Dempsey. Bensley refers to the platform.

AMD will stake its claim into this same
market by selling rebadged Athlon 64's as the Opteron 1000-series.

Sorry, I'm not seeing how selling a two socket solution, which will
undermine their marketing efforts for workstatations and servers is
relevant.

It's
all part of the two companies' counterpunching, they will not leave any
market alone that the other one has a presense in.

OK, I'll be waiting for AMD's IPF compatible processor line...and their
embedded designs.

Besides, this comes from the very foundations of the whole x86 server
industry. The whole industry got started back in the 80's, because
people were turning desktops into servers (often by simply replacing
DOS with Netware at the time). It's quite obvious that there is still a
market for such simple no-frills servers.

That's right. It's called the cheap one socket solution. How does
introducing that into the 2 socket space help anyone?

With both Intel and AMD
rebadging their desktop chips as server chips, you get to use cheap
desktop boards, with cheap desktop memory, and cheap desktop hard
disks, etc. The advantage for a customer might be that AMD and Intel
will give these processors some additional server-level warranty
support. The advantage to Intel and AMD would be that they get to
charge a little bit more for the additional support (but not too much
more).

With 4x4 motherboards out there, this might give AMD an opportunity to
open up a market not served by Intel yet -- the no-frills dual-socket
server market.

Uh huh. What market is that? Be specific? Who are the end users,
what are the distribution methods, what are the price points, what is
the level of profitability.

Try and use your head. If before you were buying Opteron based 2S
systems, and now you can buy 2S Athlon systems, where the MPUs cost
less, how is AMD going to make money?

Again this will be all desktop parts.

Right, except for that custom motherboard.

Slightly up from
the no-frills single-socket market. It'll also be different than the
professional dual-socket market, which will use Opteron 2000-series
processors (Socket F), rather than Opteron 1000 (Socket AM2). Opteron
2000 requires all of that expensive registered DIMMs crap, Opteron 1000
won't. Registered memory is useful when you need a *LOT* of memory,
but not so much if you don't.

Sorry, I just see a marketing problem. How is AMD going to avoid
eating their own lunch? What happens when everyone starts buying 4x4
systems with cheaper CPUs than Opterons?

Or worse, if Opteron systems are cheaper, what was the point of 4x4?

The secret to riding out the ups'n'downs of the cyclical market is to
have your toes dipped into every market. They're not all going to be
down at the same time.

Ok Alan Greenspan.

DK

 

[David Kanter]

If buying decisions were made on performance only, Conroe would take

the 2p market away from Opteron. However price is a big part of the
picture. If one could live without Registered memory (web servers
sure coud, ditto low and midrange workstations and some DB servers too
- such as development servers) 4x4 would offer savings too significant
to be ignored.

I don't think so. It uses a custom motherboard that will only be
produced in low volume. Low volume = high cost.

Secondly, if AMD prices their 4x4 CPUs higher than Opterons, nobody
will buy them. If they price their 4x4 CPUs lower than Opteron, then
nobody would bother buying Opteron (unless they feel they really need
ECC).

Each sale taken from Xeon is a win for AMD, even if
it's for lower margin A64 or even (cough) Sempron.
While this would
surely cannibalize some 2p Opteron as well, it is a bigger threat to
traditionally pricier Xeon that goes with yet pricier FB-DIMMs.

Uh huh. Sure it is. Have you considered the fact that AMD is going to
have to compete with all of Intel's old P4 based systems, that are
priced EXTREMELY low. While Intel's Woodcrest can demand higher
premiums, because it offers higher performance.

You really need to think these things through better. 4x4 is not going
to threaten anyone, except AMD's existing product lines. Having 4
processors is useless for gaming; there's really no way around it.
It's going to sell in ridiculously small numbers.

Intel
would have to either consede market share or drastically lower Xeon
and accompanying chipset price - both ways good for the competitor.
As it turned out, Intel cash hoard has a finite size - otherwise why
would Intel cut 10000+ heads hoping to save $3BN by 2008 - a puny
number for Intel just a few years ago. If AMD takes the price war to
2p segment, it would lighten Intel's competitive pressure in other
(desktop, mobile) segments.

Let me know when that happens. I won't be holding my breath.

4x4 would also offer some (not much but still) performance advantage
over Opteron because consumer parts were most of the time a speed
grade or two ahead of server parts, and also because unbuffered memory
is faster than buffered. If Opteron slightly trails Xeon in 2p, 4x4
may get closer in performance to Xeon, or even get ahead.

You should try to use 'reality' glasses, not 'virtual unreality' ones.
The performance gap between Woodcrest and socket F opterons is far too
large for 4x4 to make a difference. Even worse, if the workload
requires a lot of memory, 4x4 will basically be stuck going to disk.
Opteron does not 'slightly trail' Woodcrest.

Again, I suggest to reserve the final judgement on this platform, as
on any other, until after it is released. Until then any opinion,
including mine, is nothing more than speculation.

Suggest away, and I'll be ignoring you. I have yet to hear a
compelling argument for what market 4x4 addresses, how it will do so,
and how AMD will avoid shooting themselves in the foot.

DK

 

[YKhan]

Not very hard, apparently.

It's not my problem if it goes over your head. You really should read
through the whole message more often before starting your reply.

Notice how these are 1S systems.

No!? Duh!

You mean Dempsey. Bensley refers to the platform.
Fine.


Sorry, I'm not seeing how selling a two socket solution, which will
undermine their marketing efforts for workstatations and servers is
relevant.

Maybe if you read a little further down the paragraphs before replying
these problems wouldn't occur? Reading a whole message helps in
understanding things.

OK, I'll be waiting for AMD's IPF compatible processor line...and their
embedded designs.

And I'd like to see Intel's 8P+ x86 lineup too, but we're not really
going to see that are we?

That's right. It's called the cheap one socket solution. How does
introducing that into the 2 socket space help anyone?

No, it's called the cheap *desktop* solution, has nothing to do with
how many sockets there are. You get to use whatever is available in the
desktop world right now, and make use of it as a server. If desktops
are gaining dual-core and dual-socket capabilities, then you get to
turn that into a server too.

Uh huh. What market is that? Be specific? Who are the end users,
what are the distribution methods, what are the price points, what is
the level of profitability.

The market is the small business market, you couldn't figure that out
yourself? These are the businesses that may only need one or two
servers and don't have a computer room or even dedicated IT staff. The
price points are whatever you can buy as a desktop with a slightly
better warranty.

Try and use your head. If before you were buying Opteron based 2S
systems, and now you can buy 2S Athlon systems, where the MPUs cost
less, how is AMD going to make money?

I'm not the one having a problem using my head here.

Because there are businesses who need enterprise-grade features and
then there are businesses who don't. The enteprise grade features would
include registered and ECC DIMMs, rack-mounting form-factors, redundant
power supplies, large amounts of RAM, RAIDed disks, 24/7 guaranteed
availability, among other things. Some of these things might be useful
for small businesses, but usually they're not all necessary.

Right, except for that custom motherboard.

I doubt anyone would categorize an AMD 4x4 motherboard as "custom".

Sorry, I just see a marketing problem. How is AMD going to avoid
eating their own lunch? What happens when everyone starts buying 4x4
systems with cheaper CPUs than Opterons?

This has been answered above. To repeat, some people need enterprise
features, some people don't.

Or worse, if Opteron systems are cheaper, what was the point of 4x4?

An Opteron 1000 is still an Opteron system.

Yousuf Khan