Intel releases dual-core Xeon finally

  • Thread starter Thread starter YKhan
  • Start date Start date
EdG said:
Are these like the DC Pentium 4's, just 2 chips slapped together?

Worse, they're just two seperate chips sitting side by side a few
microns apart inside the same package.

For some reason this is a higher performance option for Intel than its
Pentium D Smithfield method of two fused Pentium 4 cores. For everyone
else, having the two cores on the same die usually works out faster,
but not for Intel. :-)

Yousuf Khan
 
Worse, they're just two seperate chips sitting side by side a few
microns apart inside the same package.

For some reason this is a higher performance option for Intel than its
Pentium D Smithfield method of two fused Pentium 4 cores. For everyone
else, having the two cores on the same die usually works out faster,
but not for Intel. :-)

Paxville has 2M L2 per core, so it can't be Smithfield, which is 1M/core. I
assume it's 2 of the 6xx series P4's bolted together, and they either
didn't have time to do it on one die, or decided it wouldn't be economical.
 
EdG said:
Are these like the DC Pentium 4's, just 2 chips slapped together?
As opposed to what? What else would a dual core be other than two cores
joined in a single carrier?

You can make cases for several ways to joind them, and various cache
sharing (or not) methods, but in the long run you still have two cores
slapped together.

If Intel wanted to impress me, they could release a CPU which would drop
in place of the one I have, and with only a BIOS flash I could be
running faster. They didn't impress me, clearly there are reasons both
technical and financial why that isn't happening.

I wonder how Intel differentiates the Xeon from the DC-EE?
 
EdG said:
Are these like the DC Pentium 4's, just 2 chips slapped together?

Oh and here's some hilarity, this article takes the stance that Intel
dual-core will now "leapfrog" AMD. Of course they take a very liberal
definition of "leapfrog". To them, leapfrog means that Intel will come
into parity with today's AMD offerings, in about a year's time. :-)

Intel Could 'Leapfrog' AMD With New Processors - Forbes.com
http://www.forbes.com/markets/2005/10/11/intel-amd-processors-1011markets02.html?partner=yahootix

Yousuf Khan
 
YKhan said:
Worse, they're just two seperate chips sitting side by side a few
microns apart inside the same package.

For some reason this is a higher performance option for Intel than its
Pentium D Smithfield method of two fused Pentium 4 cores. For everyone
else, having the two cores on the same die usually works out faster,
but not for Intel. :-)

I can't think of any killer argument for either choice. There are all
sorts of manufacturing and testing issues, and maybe "faster" means time
to market rather than performance. I would expect using a single
substrate would be faster, but I'm clearly not a chip designer. And if
someone has some insight on this I'd like to hear it.
 
As opposed to what? What else would a dual core be other than two cores
joined in a single carrier?

Two cores - one chip.
You can make cases for several ways to joind them, and various cache
sharing (or not) methods, but in the long run you still have two cores
slapped together.

One chip is not "slapped together", in as much as "two chips on a
single carrier" is.
If Intel wanted to impress me, they could release a CPU which would drop
in place of the one I have, and with only a BIOS flash I could be
running faster. They didn't impress me, clearly there are reasons both
technical and financial why that isn't happening.

You mean like AMD? ;-)
I wonder how Intel differentiates the Xeon from the DC-EE?

Many $$. ;-)
 
Techworld.com - Intel's dual-core Xeon opens a new chapter
http://www.techworld.com/opsys/features/index.cfm?featureid=1854&inkc=0

Paxville is a boutique Xeon part for servers, in the model of the EE series
P4s for the desktop. It enables reuse of Lindenhurst chipsets and motherboards
(even if you didn't see this coming, it's only two wires and four resistors),
and will have a short lifespan (not clear if there'll even be a speed-bump
before they hit EOL).

We've been running them for a few months with our Lindenhurst-VS designs quite
successfully, and expect to complete our internal certification before we
roll-out products with 'em in early aught-six.

Upside: they do kick some benchmark booty - as long as the benchmarks are
cache-friendly.

Downside: Prodigious Power & Cooling requirements. We had to uprate the
loadlines on our VRDs to deal with the power piece, which thereby required
uprated components, and now our thermal guys are pulling their hair out to
deal with the cooling piece without pissing off the acoustics guys, who have
the regulatory guys waiting with drawn knives...

/daytripper (just another day in Dilbertville ;-)
 
Worse, they're just two seperate chips sitting side by side a few
microns apart inside the same package.

For some reason this is a higher performance option for Intel than its
Pentium D Smithfield method of two fused Pentium 4 cores. For everyone
else, having the two cores on the same die usually works out faster,
but not for Intel. :-)

Yousuf Khan

Then expect the same packaging issues that dogged P-Pro. Double the
risk of getting a dude in the end. While testing and packaging
technology improved quite a bit since, and Xeon volume is orders of
magnitude less than that of Pentium, yet there will be quite a few
good cores trashed because the core next door is iffy. But anyway
now, half a year afted dual core Opterons came out, Intel and its
faithful Dell can finally say "Me too".
NNN
 
YKhan said:
Worse, they're just two seperate chips sitting side by side a few
microns apart inside the same package.

For some reason this is a higher performance option for Intel than its
Pentium D Smithfield method of two fused Pentium 4 cores. For everyone
else, having the two cores on the same die usually works out faster,
but not for Intel. :-)

Um, it's in fact ALWAYS higher clockspeed. If you go the 1 die for 2
CPUs, then the clockspeed will be the minimum of the two.

If you go two dice on a package, then you can bin the dice
independently and mix & match the appropriate clockspeed ones. Of
course, it isn't quite as good at dealing with CC issues as if you
shared the caches...

David
 
Um, it's in fact ALWAYS higher clockspeed. If you go the 1 die for 2
CPUs, then the clockspeed will be the minimum of the two.

....and they will track rather closely.
If you go two dice on a package, then you can bin the dice
independently and mix & match the appropriate clockspeed ones. Of
course, it isn't quite as good at dealing with CC issues as if you
shared the caches...

Not really. It's very hard to do final test before the dice are
packaged. Either way, two dice are more expensive than one.
 
Yousuf said:
Oh and here's some hilarity, this article takes the stance that Intel
dual-core will now "leapfrog" AMD. Of course they take a very liberal
definition of "leapfrog". To them, leapfrog means that Intel will come
into parity with today's AMD offerings, in about a year's time. :-)

Intel Could 'Leapfrog' AMD With New Processors - Forbes.com
http://www.forbes.com/markets/2005/10/11/intel-amd-processors-1011markets02.html?partner=yahootix


Yousuf Khan

Boy is that bad reporting! These three quotes came from the same
article above.

Intel Could 'Leapfrog' AMD With New Processors

UBS said that Intel's next-generation 65nm processors "provide a major
leap in performance" and that even after AMD migrates to 65nm
processors, "we expect Intel to have a performance lead in DP
[dual-processor] servers."

we expect Truland to lag AMD's Opteron system by about 10% on TPC

Taking a quick stroll to Tom's Hardware, which has a review up from a
chip-in-hand. The 65nm Pentium 4 runs at lower power than 90nm Pentium
4, but has exactly the same performance (<0.1% difference doesn't amount
to anything). If 90nm Pentium 4 is considered well behind 90nm Athlon
64 (I'm pretty sure that's still the consensus out in internetworld),
then 65nm Pentium 4 can't be ahead. Why do they torture us this way?
Hurry up and get the dang Pentium M on desktops! Focus, guys, focus!

Alex
 
As opposed to what? What else would a dual core be other than two cores
joined in a single carrier?

You can make cases for several ways to joind them, and various cache
sharing (or not) methods, but in the long run you still have two cores
slapped together.

If Intel wanted to impress me, they could release a CPU which would drop
in place of the one I have, and with only a BIOS flash I could be
running faster. They didn't impress me, clearly there are reasons both
technical and financial why that isn't happening.

AMD systems seem to have Intel beat hands down as far as getting life
out of the chips and boards, every time I wanted to upgrade my Intel it
was,.... oh the bios doesn't support that chip, or you need a diff
chipset, or Intel change the voltage on the new CPUs', or it's a full
moon, bla bla bla, was always something! The 3rd time I got I burned I
went out and bought AMD, been buying AMD ever since, Intel is just a
well known brand, their nothing special, they remind me of over-priced
Sony crap , paying for a name, not the product.
I wonder how Intel differentiates the Xeon from the DC-EE?

Isn't Intel supposed to have an all new desktop platform out by mid 2006
that's supposed to blow AMD away? I read something about AMD doing a
new socket in 06 too, but is that just for Opteron?

Cheers,
Ed
 
Yousuf said:
Oh and here's some hilarity, this article takes the stance that Intel
dual-core will now "leapfrog" AMD. Of course they take a very liberal
definition of "leapfrog". To them, leapfrog means that Intel will come
into parity with today's AMD offerings, in about a year's time. :-)

Intel Could 'Leapfrog' AMD With New Processors - Forbes.com
http://www.forbes.com/markets/2005/10/11/intel-amd-processors-1011markets02.html?partner=yahootix


Yousuf Khan

Boy is that bad reporting! These three quotes came from the same
article above.

Intel Could 'Leapfrog' AMD With New Processors

UBS said that Intel's next-generation 65nm processors "provide a major
leap in performance" and that even after AMD migrates to 65nm
processors, "we expect Intel to have a performance lead in DP
[dual-processor] servers."

Ya and they all bought Intel's hype about how the P4 was designed to
last Intel 10 years, and would be running at 10GHz by 2009. Intel talks
a lot shit!
 
Keith said:
Two cores - one chip.

As noted below in my first post, there are several ways to do this,
sharing substrate or not, sharing some level of cache, the differences
being in how well some subset of applications runs due to the choice. At
the end user level I think it just boils down to two CPU operation from
one package, the details are not part of the buying choice, not do they
need to be.
One chip is not "slapped together", in as much as "two chips on a
single carrier" is.

As opposed to slapped together at the mask. I take your point, but I'm
still lacking a great argument that one method is better than the other.
Both matching chip speed before bonding the the substrate and cache
sharing have been mentioned, the problem is that they're exclusive
advantages. And no one has shipped a chip with bonded multi-core and a
separate cache, L2 or L3, also bonded. Intel did CPU and cache as two
bonded chips, so there's some prior art.
You mean like AMD? ;-)

Is there really a motherboard vendor who has shipped a dual-core BIOS
fix for a board designed and sold for single core operation?

I'm told that a dual core Intel EE chip can be dropped into a recent HT
aware motherboard and that the voltages are compatible. Since I can't
get a BIOS which will work, I'm not motivated to investigate.
Many $$. ;-)

The question was serious, why would anyone buy a Xeon at this point?
 
Bill said:
As noted below in my first post, there are several ways to do this,
sharing substrate or not, sharing some level of cache, the differences
being in how well some subset of applications runs due to the choice. At
the end user level I think it just boils down to two CPU operation from
one package, the details are not part of the buying choice, not do they
need to be.

As opposed to slapped together at the mask. I take your point, but I'm
still lacking a great argument that one method is better than the other.
Both matching chip speed before bonding the the substrate and cache
sharing have been mentioned, the problem is that they're exclusive
advantages. And no one has shipped a chip with bonded multi-core and a
separate cache, L2 or L3, also bonded. Intel did CPU and cache as two
bonded chips, so there's some prior art.

Is there really a motherboard vendor who has shipped a dual-core BIOS
fix for a board designed and sold for single core operation?

Tyan. I had to install such a fix for several S2885 (dual socket
940) boards in order to upgrade some systems from Opty 24x to
Opty 270.

Without the BIOS upgrade, they wouldn't even POST.

MSI also has similar BIOS upgrades available for download, but I
haven't personally tried them.
 
As noted below in my first post, there are several ways to do this,
sharing substrate or not, sharing some level of cache, the differences
being in how well some subset of applications runs due to the choice. At
the end user level I think it just boils down to two CPU operation from
one package, the details are not part of the buying choice, not do they
need to be.

The cache architecture is irrelevant dual-core <> two processors on one
substrate. Plunking two chips on a substrate is *expensive* and in *all*
ways sub-optimal.
As opposed to slapped together at the mask. I take your point, but I'm
still lacking a great argument that one method is better than the other.

Please! Think interconnect. Think cost! Think test, fer chrissake!
Both matching chip speed before bonding the the substrate and cache
sharing have been mentioned, the problem is that they're exclusive
advantages. And no one has shipped a chip with bonded multi-core and a
separate cache, L2 or L3, also bonded.

What the hell are you saying? Sheesh!
Intel did CPU and cache as two bonded chips, so there's some prior art.

Of course there is. It's *EXPENSIVE*! IBM made 121 chip modules in the
80s. Guess what, they were *EXPENSIVE*. Yikes!

IBM has been making 10 CPU modules (with dual core chips) since the
POWER-4 line, with integrated L1&L2, and on-substrate L3. So what?
The packaging is *EXPENSIVE*.
Is there really a motherboard vendor who has shipped a dual-core BIOS
fix for a board designed and sold for single core operation?

Of course. Do try to keep up.
I'm told that a dual core Intel EE chip can be dropped into a recent HT
aware motherboard and that the voltages are compatible. Since I can't
get a BIOS which will work, I'm not motivated to investigate.

I'm impressed. They still have two chips, with two loads on the bus.
Whoopie!
The question was serious, why would anyone buy a Xeon at this point?

I am too. Indeed, why would anyone buy Intel?
 
Back
Top