IA64 to be dual core by 2005
JF Mezei
(sorry if it wraps, essential text below:)
SAN JOSE, Calif. (Reuters) - Intel Corp. the world's largest chip maker, on
Tuesday outlined plans for two new chips that will have two or more processors
on a single piece of silicon, boosting the performance of corporate server
computers.
The new chips are a 32-bit Xeon server processor MP, code-named "Tulsa," which
will be its first dual-core chip, and a new 64-bit Itanium server chip,
code-named "Tanglewood," Intel President Paul Otellini said in his keynote
address at the Intel Developer Forum.
Tanglewood will come some time after 2005 and Tulsa in two to three years, he
said.
In an interview, Otellini declined to say what plans Intel has for bringing
64-bit chips to PCs. Currently, its chips for PCs crunch 32 bits of data at a
time.
"The production operating systems are not there yet" for 64-bit desktop
computers, he said. "The mainstream applications won't exist until next year."
Apple Computer Inc. already offers a personal computer based on a 64-bit
chip, while Intel rival Advanced Micro Devices Inc. is set to unveil its
Athlon 64 for PCs later this month.
A future Itanium processor, code-named "Montecito," will be the first chip
with one billion transistors, according to Otellini. It is targeted for
production in 2005, he said.
Otellini also gave a timeline for pending chip manufacturing process
technologies that allow Intel to shrink the circuitry and space between
transistors to fit more of them onto a chip, thus increasing their computing
power with no increase in cost or size.
He showed for the first time a wafer containing chips made with 65 nanometer
process, which will be in production in 2005, he said.
Those chips will be followed by 45 nanometer process chips in 2007; 32
nanometer process chips in 2009; and, in 2011, 22 nanometer process chips,
which will have transistor space that is smaller than the width of a DNA
molecule, he said.
JF Mezei
what does dual core mean from the point of view of the operating system and
scheduling ?
Does it appear as a single (faster) chip to the OS, or does the OS see 2 cpus
which it can use for SMP etc ?
John Smith
http://www.eweek.com/article2/0,4149,1269660,00.asp
Intel CTO: Desktop Isn't Ready for 64 Bits
By Mark Hachman
September 15, 2003
SAN JOSE, Calif.-Sixty-four bit computing won't be needed on desktop
PCs for several more years, according to Pat Gelsinger, chief technology
officer at Intel Corp. ......
ie. no volume in the 64-bit market from Intel for years.....read
higher prices....read lower adoption rates...
Ryan Moore
Assuming it's the same sort of technology as the new P4 "hyperthreading",
then it would show up as two CPUs on one piece of silicon. In the case of
P4 "hyperthreading", there are two sets of execution units, but some other
components are shared. Thus, the CPU isn't as fast as two complete CPUs.
The more componenets that are shared between the cores, the less speed-up
that can be accomplished.
VMS would see two CPUs and would do SMP scheduling. There would probably
need to be support in the OS to do it, though.
-Ryan
hea...@aracnet.com
> On Tue, 16 Sep 2003, JF Mezei wrote:
> > question:
> >
> > what does dual core mean from the point of view of the operating system and
> > scheduling ?
> >
> > Does it appear as a single (faster) chip to the OS, or does the OS see 2 cpus
> > which it can use for SMP etc ?
> Assuming it's the same sort of technology as the new P4 "hyperthreading",
> then it would show up as two CPUs on one piece of silicon. In the case of
> P4 "hyperthreading", there are two sets of execution units, but some other
> components are shared. Thus, the CPU isn't as fast as two complete CPUs.
Dual Core refers to two seperate CPU's on the same chip. I believe IBM has
been doing this for at least a year or two with POWER4. Basically it lets
you have a true dual processor system with only one chip.
Zane
Rob Young
2 CPUs seen as one. Something I speculated on here:
And the same info quoted from news.com back to comp.os.vms (this
article:
http://groups.google.com/groups?selm=d7791aa1.0301181143.23c3574c%40posting.google.com&output=gplain
From: b...@instantwhip.com (Bob Ceculski)
Newsgroups: comp.os.vms
Subject: Itanium ahead of itself, not behind ...
Date: 18 Jan 2003 11:43:39 -0800
Besides enhancing performance, Intel may use its dual-core chips to
undercut IBM, said Kevin Krewell, senior editor at the Microprocessor
Report. In larger servers, Oracle and other software vendors charge a
licensing fee for every processor in a given server.
To date, IBM has said that the Power4, although a single chip, has two
processors. Hence, software customers have to buy two licenses for
each Power4 chip. >>> INTEL IS ALREADY INDICATING THAT IT WILL CONSIDER
MONTECITO A SINGLE PROCESSOR, REQUIRING ONLY ONE SOFTWARE LICENSE,
KREWELL SAID. <<<
"I think they are going to sell it that way to make it cost effective"
to switch to Itanium from other servers with different chips, Krewell
said.
Yesterday:
http://www.theinquirer.net/?article=11579
The big announcement is Vanderpool, or Vanderpool Technology (VT), which is
Intel's way of putting VMWare functionality on a chip. The time for hardware
partitioning is near, or at least the time for more announcements about
hardware partitioning is near, they didn't put a time frame on this technology
family. [How about 2005 with Monticeto?]
In short, it allows you to run two virtual CPUs on a single CPU, basically like
having a second PC in the box. This is not HT technology, which emulates a
second CPU, the demo they showed had one 'computer' playing a Simpsons clip,
and the other playing a game, rebooting, and then installing drivers. Nothing
that you can't do now in software, the take home message is that this
technology will decrease, or possibly eliminate the overhead of this
increasingly important corporate mainstay.
---
The advantages are huge as I speculated earlier:
There is a huge incentive.
http://www.intel.com/technology/itj/2002/volume06issue01/art03_specprecomp/p05_hw_vs_sw.htm
"Initial results indicate that the prefetching performed by the speculative
threads can achieve significant speed-ups on an in-order processor, ranging
from 16% to 104%, on pointer-intensive benchmarks."
Imagine if it sped up Oracle/MSQL 40-80% per Itanium
CPU without increasing the external
CPU count thereby keeping
down expensive licensing costs. Just a thought.
---
Speculation:
Oracle WILL be redoing their licensing. However, those $40000
per-CPU licensing costs that you pay for today on 8, 9i and 11
will become a WHOLE lot better with Monticeto. The ROI may
be: swap out your current box for Monticeto and save money in
the process. Seems that will be the case (for the common case -
i.e. running out of horsepower in that 4-cpu box - get a 4-cpu
Montecito which provides 50-70% higher performance).
Rob
Rob Young
> Yesterday:
>
> http://www.theinquirer.net/?article=11579
>
> The big announcement is Vanderpool, or Vanderpool Technology (VT), which is
> Intel's way of putting VMWare functionality on a chip. The time for hardware
> partitioning is near, or at least the time for more announcements about
> hardware partitioning is near, they didn't put a time frame on this technology
> family. [How about 2005 with Monticeto?]
>
> In short, it allows you to run two virtual CPUs on a single CPU, basically like
> having a second PC in the box. This is not HT technology, which emulates a
> second CPU, the demo they showed had one 'computer' playing a Simpsons clip,
> and the other playing a game, rebooting, and then installing drivers. Nothing
> that you can't do now in software, the take home message is that this
> technology will decrease, or possibly eliminate the overhead of this
> increasingly important corporate mainstay.
>
Vanderpool technology is a ways off so how Montecito appears
as a single CPU is still a bit of a mystery.
http://asia.reuters.com/newsArticle.jhtml?type=technologyNews&storyID=3454786
Vanderpool, which will be available in the next five years, will create a
partition inside the chip to allow it to be used for different programs on
different machines. For instance, Otellini demonstrated a chip with Vanderpool
technology that allowed him to watch a TV program on a TV while another person
was playing a video game.
Montecito uses an "arbiter" but how that works or what it does is
speculation (maybe it is still doing speculative computation - but
that is a guess and a high risk of being totally wrong and maybe
not nearly the time to verify and stuff it into silicon):
http://www.eetimes.com/semi/news/OEG20030210S0007
Then, in 2005, Intel will unveil the dual-core Montecito. Each core will have a
processor and L3 cache in the same package. Intel has also developed what is
calls an "arbiter" bus technology, which will manage two cores as one systems
bus.
The "arbiter" itself appears to be an internal bus that will interface to the
overall systems bus. The "arbiter" technology will enable a processor to
support twice the cache over previous devices, Modi said.
Little else was disclosed about the "arbiter" but the technology is key for
high-end servers. "The fundamental driver is the cache," he said. "You get
significantly better latencies with more on-die cache."
The "arbiter" will support the dual-core Montecito, but Intel dropped hints
that it may be required to handle multi-core processors. "We are not ruling out
multi-core systems in the future," he said.
---
Maybe the whole business of appearing as a single CPU is
wrong as Bill pointed out before.
Rob
Fred Kleinsorge
news:3F6785FD...@istop.com...
Dual core looks exactly like 2 CPUs in a single CPU socket. So it allows
2-socket systems to become 4-way, 4-socket to become 8-way, etc.
Fred Kleinsorge
"Ryan Moore" <rmo...@rmoore.dyndns.org> wrote in message
news:Pine.LNX.4.56.0309161906170.6107@jaipur...
Look mom, no hands! Dual core for the most part will be transparent to the
OS, and just doubles the CPU capacity.
JF Mezei
> each Power4 chip. >>> INTEL IS ALREADY INDICATING THAT IT WILL CONSIDER
> MONTECITO A SINGLE PROCESSOR, REQUIRING ONLY ONE SOFTWARE LICENSE,
> KREWELL SAID. <<<
OK then. As a single processor, what is the advantage of a dual core ? If they
insist on appearing as a single CPU to the OS, shouldn't they then focus on
adding more execution pipelines (or whatever the terminology is) ? What is the
point of a dual core if the second core cannot act independantly as a separate
CPU ?
> "I think they are going to sell it that way to make it cost effective"
> to switch to Itanium from other servers with different chips, Krewell
> said.
I think it is easy for Intel to talk like this now since their dual core stuff
is so far away. It may be much harder for them to be cost competitive if a
single CPU costs a lot more because it has a dual core but you're not allowed
to use the second core.
> Intel's way of putting VMWare functionality on a chip. The time for hardware
> partitioning is near, or at least the time for more announcements about
> hardware partitioning is near, they didn't put a time frame on this technology
> family. [How about 2005 with Monticeto?]
Well, this is no surprise since existing chips/systems such as Alpha already
have the ability to be building blocks of systems that support partitioning.
So Intel has to add that functionality to its IA64 if IA64 is to survive and
replace Alpha and PaRisc.
However, where partitioning woudl be needed is really in the wintel world so
that you could run multiple instances of Windows on a single big box, allowing
you to minimize the hardware and still have one instance of windows per
application to minimise the impact of windows hanging.
JF Mezei
> Look mom, no hands! Dual core for the most part will be transparent to the
> OS, and just doubles the CPU capacity.
Are you talking about Intel's specific plans for IA64, or is that a general
statement applicable to existing platforms such as Power etc ?
JF Mezei
>Look mom, no hands! Dual core for the most part will be transparent to the
>OS, and just doubles the CPU capacity.
Fred Kleinsorge wrote in another post:
> Dual core looks exactly like 2 CPUs in a single CPU socket. So it allows
> 2-socket systems to become 4-way, 4-socket to become 8-way, etc.
OK, which is which ? Will the OS see a single CPU or will it see multiple CPUs
allowing SMP etc etc ?
Rob Young
> Rob Young wrote:
>> each Power4 chip. >>> INTEL IS ALREADY INDICATING THAT IT WILL CONSIDER
>> MONTECITO A SINGLE PROCESSOR, REQUIRING ONLY ONE SOFTWARE LICENSE,
>> KREWELL SAID. <<<
>
> OK then. As a single processor, what is the advantage of a dual core ? If they
> insist on appearing as a single CPU to the OS, shouldn't they then focus on
> adding more execution pipelines (or whatever the terminology is) ? What is the
> point of a dual core if the second core cannot act independantly as a separate
> CPU ?
>
<Backpedling mode>
As Bill pointed out in realworldtech.com when this discussion
came up a few weeks ago ... Krewell may have it wrong. I can't
find definitive evidence of what Montecito's "arbiter" does, if
Montecito does indeed have Speculative Precomputation or any
mechanism to "appear as a single CPU" to the OS.
>> "I think they are going to sell it that way to make it cost effective"
>> to switch to Itanium from other servers with different chips, Krewell
>> said.
>
> I think it is easy for Intel to talk like this now since their dual core stuff
> is so far away. It may be much harder for them to be cost competitive if a
> single CPU costs a lot more because it has a dual core but you're not allowed
> to use the second core.
>
It isn't far away at all. 2005 is 15 months away - maybe 19 months
if they ship Montecito Q2 2005.
>> Intel's way of putting VMWare functionality on a chip. The time for hardware
>> partitioning is near, or at least the time for more announcements about
>> hardware partitioning is near, they didn't put a time frame on this technology
>> family. [How about 2005 with Monticeto?]
>
> Well, this is no surprise since existing chips/systems such as Alpha already
> have the ability to be building blocks of systems that support partitioning.
> So Intel has to add that functionality to its IA64 if IA64 is to survive and
> replace Alpha and PaRisc.
VMWare or Vanderpool appears to be across CPU lines. Perhaps
Intel steps up and explains it or that info shows up
on the WibblyWeb.
> However, where partitioning woudl be needed is really in the wintel world so
> that you could run multiple instances of Windows on a single big box, allowing
> you to minimize the hardware and still have one instance of windows per
> application to minimise the impact of windows hanging.
It was late and I was winging on - on reinspection I think
Vanderpool technology is LPARs without the LPAR - meaning
virtual CPUs carved out of 1 or more cores, if I am reading
this correctly:
http://www.internetnews.com/dev-news/article.php/3078291
Intel developers also are working on a separate technology code-named
"Vanderpool," a new performance enhancement that enables multiple, independent
software environments in a single PC. As Otellini described it, this technology
enables users to partition their home computers just as a network administrator
might partition a mainframe-class system, increasing reliability and speeding
the ability to recover from computer crashes.
So a number of technologies are baking.
1) Multiple cores that appear as one - Maybe Montecito?
Advantage: Decreased licensing costs.
2) Vanderpool technology - single CPU appear as multiple
(or virtual CPUs or systems?) such that processes run in
a separate domain.
Advantage: Application crash - just that application
goes down, others continue to run.
Questions abound. How does an OS take advantage of Vanderpool?
Is it transparent?
Rob
Fred Kleinsorge
doubling the CPU capacity, I mean that the same number of "sockets" allows
you to effectively have twice the CPU count - doubling a machines CPU
capacity.
Now, how Intel views the *chip* is something different. From their
pricing/licensing/etc pov, they may view the part as a single CPU chip.
How an OS will treat it is something different. *As* may be how an OS does
license checks -- which is a policy issue.
"JF Mezei" <jfmezei...@istop.com> wrote in message
news:3F687DDD...@istop.com...
Nic Clews
>
...
> Questions abound. How does an OS take advantage of Vanderpool?
> Is it transparent?
Leave it at "questions abound".
The IPF is also about application redesign, a new way of looking at
processing data.
The O/S provides the ability to use the hardware.
The application then has to be able to keep a parallel, or multiple
parallel processors busy. Studying the Beowulf cluster concepts and
applications may be helpful.
--
Regards, Nic Clews a.k.a. Mr. CP Charges, CSC Computer Sciences
nclews at csc dot com
Andrew Harrison SUNUK Consultancy
If you don't have a product .................
regards
Andrew Harrison
Andrew Harrison SUNUK Consultancy
I think one of the questions was how this plays with
something like Oracle that likes to know (for licensing
reasons) how many CPU's a system has).
Your answer appears to be 2 cores 2 CPU's hence 2 Oracle
licenses.
Regards
Andrew Harrison
Fred Kleinsorge
"Andrew Harrison SUNUK Consultancy" <Andrew_No....@nospamn.sun.com>
wrote in message news:bkcerf$oia$2...@new-usenet.uk.sun.com...
> Fred Kleinsorge wrote:
>
> I think one of the questions was how this plays with
> something like Oracle that likes to know (for licensing
> reasons) how many CPU's a system has).
>
> Your answer appears to be 2 cores 2 CPU's hence 2 Oracle
> licenses.
>
That is an Oracle business practices question.
JFmezei
> > Your answer appears to be 2 cores 2 CPU's hence 2 Oracle
> > licenses.
> >
>
> That is an Oracle business practices question.
I don't see it as an Oracle issue. It is a simple issue: does the CPU appear
as one CPU or 2 CPUs to the software ?
Now, if dual CPUs become common and do appear as 2 CPUs, then yes, perhaps
those companies who price their products according to number of CPUs may have
to change their business practices.
Fred Kleinsorge
"JFmezei" <jfmezei...@istop.com> wrote in message
news:3F69F2DB...@istop.com...
OK. This is all speculation since I have not yet seen dual core chips.
My understanding is they will look simply like 2 CPUs to us. They just
happen to live in the same socket. Will we be able to determine this? I
think so, but am not certain. Will we do anything special about it?
Probably not. If we only report half the CPUs, then something is certain to
be unhappy. Will we have some additional system information that someone
could find out the number of *sockets* - if it's possible to find out, we'll
probably try to provide the information through some lexical or system cell
or call.
Then it really is a business practice issue. Since I don't really think
that we should "under report" the number of CPUs in any current programmatic
query. So, to implement "per-socket" licensing would probably require
changes to SW that actually directly looks at the CPU count.
So it really is a business practices issue. SW that currently does per-CPU
licensing, and checks the # of CPU's - needs to determine if it really
should do per-Socket licensing.
John Santos
Intel said the exact same thing about Pentiums when they were introduced.
"Pentiums are for servers. Desktop users should stick with 486." they
said. We all know how long that lasted. The only diffence was there
then there was a lot of clamor from users (mostly gamers) for
Pentium desktops. I don't know how much current gamers care about
IPF, AMD-64, or 64 bits in general.
--
John Santos
Evans Griffiths & Hart, Inc.
781-861-0670 ext 539
Andrew Harrison SUNUK Consultancy
No
Regards
Andrew Harrison
Paul Repacholi
> "Pentiums are for servers. Desktop users should stick with 486."
> they said. We all know how long that lasted. The only diffence was
> there then there was a lot of clamor from users (mostly gamers) for
> Pentium desktops. I don't know how much current gamers care about
> IPF, AMD-64, or 64 bits in general.
Does the Pentium 4EE answer that question? A 2MB cache Xeon in pentium
clothing. Good for and extra 2-20% performance.
BTW, Dual is for large values of 2! Expect about 7x performance if
that gives you a hint. This is for Tanglewood I think it's called,
The mongosizler follow on.
--
Paul Repacholi 1 Crescent Rd.,
+61 (08) 9257-1001 Kalamunda.
West Australia 6076
comp.os.vms,- The Older, Grumpier Slashdot
Raw, Cooked or Well-done, it's all half baked.
EPIC, The Architecture of the future, always has been, always will be.
Bob Koehler
> Fred Kleinsorge wrote:
>>
>> That is an Oracle business practices question.
>>
>>
>
> No
What Oracle charges for any computer, and why, is an Oracle business
practices question.
Andrew Harrison SUNUK Consultancy
ohh dear thanks for stating the obvious !
Of course what Oracle charges for a computer is their business
however what Oracle currently charge on is License x Number of CPU's.
So if you have a die with 2 cores that appears to the OS and Oracle
as 2 CPU's then you incure a License x 2 x the number of dies cost.
If you have a dies with 2 cores that appears to the OS and Oracle
as 1 CPU then you incure a License x 1 x nubmer of dies cost.
Now for the tricky bit (apparently).
Whether the each die appears as 1 or 2 CPU's to the OS and Oracle is
entirely a technical issue.
I hope this clears up your apparent confusion, let me know if it
doesn't.
regards
Andrew Harrison
Bob Koehler
>
> ohh dear thanks for stating the obvious !
If it was so obvious, then why have you been disagreeing with it?
> Of course what Oracle charges for a computer is their business
> however what Oracle currently charge on is License x Number of CPU's.
Since they're not yet dealing with dual core chips, their current
practices don't predict how they will deal with them when they come
out.
Andrew Harrison SUNUK Consultancy
Ohhhhhhhhhhh Yeeeeeessssssssss Theeeeeeeeeeeeeeey Areeeeeeeeeeeeee
How about Power 4 Dual core now
Xeon Sort of dual core now, appears as 2 CPU's
For Power 4 oracle currently charge as if the die has
two CPU's unless its in a P690 HPC which has 1/2 the
cores disabled to allow each core to have 2x L2 and 2xL3
cache, they then charge for 1 CPU per die.
So there is a model currently which is dual core which
the OS sees as 2 CPU's = 2 x the licenseing costs.
regards
Andrew Harrison
JF Mezei
> Since they're not yet dealing with dual core chips, their current
> practices don't predict how they will deal with them when they come
> out.
OK, lets get this closer. Do VMS licences cost more if you have more than one
cpu in a machine ?
Would an OS or application see a difference between 2 separate CPUs and one
chip that contains 2 CPUs inside ?
If dual code CPUs become common, perhaps software vendors will have to change
their business practice to charge the same for 1 or 2 CPUs and then milk
customers when they go above that.