[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference ricks::dechips

Title:Hudson VLSI
Notice:For Digital Chip Data - CHIPBZ::PRODUCTION$:[DS_INFO...]
Moderator:RICKS::PHIPPS
Created:Wed Feb 12 1986
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:701
Total number of notes:4658

678.0. "Pentium II news" by GEMGRP::PIEPER () Fri Apr 11 1997 14:29

from http://www.intel.com/pressroom/archive/releases/CN040997.HTM
===========


Intel�s CEO Reveals New Bus Architecture To Be Implemented In
Upcoming Pentium� II Microprocessor

New Implementation Addresses Bandwidth "Valleys of Death"

HOUSTON, Texas, April 9, 1997 � Intel Corporation CEO Andrew S. Grove
outlined today that the Dual Independent Bus (DIB) architecture will be
implemented in the Pentium� II microprocessor, which will be formally
introduced in May. 

Speaking to an audience of information technology managers at Compaq's
"Innovate" conference, Grove described how the Dual Independent Bus
architecture will dramatically improve the ability of the core processor to
exchange data with memory subsystems. "Intel's microprocessor performance
continues to increase at a phenomenal rate. The Dual Independent Bus
architecture allows total system performance to scale with microprocessor
performance." Grove said. 

Valleys of Death

Dr. Grove described the bandwidth bottlenecks in current PC platforms as
bandwidth "valleys of death" that limit the output and performance of the
microprocessor. "Processor performance alone is not enough," he told the
attendees at Compaq Computer Corp.�s Innovate conference. "We must address
two �valleys of death,� namely, the processor-to-memory and
processor-to-graphics bus bandwidth bottlenecks." 

The introduction of Pentium II processor systems later this quarter with
the DIB architecture and the addition of a new feature called the
Accelerated Graphics Port (AGP) later this year will address these
bandwidth bottlenecks. Collectively, these technologies will provide
performance and bandwidth that will scale with Intel processors, which are
expected to reach clock rates beyond 500 MHz by the year 2000. 

The Visual Connected PC Vision Demonstrated

Grove then went on to introduce the next step in business computing, the
Visual Connected PC. The Visual Connected PC consists of a minimum of a
Pentium II processor with its Dual Independent Bus architecture, MMX�
technology, broad use of the Internet, manageability capabilities, graphics
through AGP, and 100 Mbit Ethernet networking interface. 

Throughout his presentation, Grove demonstrated the visual computing
capabilities as well as the manageability features of a wide range of
visual, connected PCs. These technology demonstrations included: 

          A 300 MHz Pentium II processor workstation performing solid modeling. 
          Real-time 3D rendering on both a 300 MHz and 500 MHz Pentium II
          processor-based workstation 
          Web-based manageability of Net PCs using LANDesk� Configuration
          Manager. 
          A 266 MHz Pentium II processor-based NetPC running Java applications
          natively. 
          A 166 Mhz Pentium Processor with MMX technology-based notebook being
          utilized to transact business visually over the Internet.

Dual Independent Bus Architecture = Improved Memory Bandwidth Performance

Grove also provided technical details on how greater bandwidth will be
achieved. The Dual Independent Bus architecture was first implemented in
the Pentium Pro processor and will continue with the Pentium II processor.
The bus architecture dramatically improves the ability of the core
processor to exchange data with the memory subsystems over processors with
a single bus system like the Pentium processor. 

Two buses make up the Dual Independent Bus Architecture: the L2 cache bus
and the processor-to-main-memory system bus. The single dedicated L2 cache
on the Pentium II processor operates twice as fast as the L2 cache on a
Pentium processor. The pipelined system bus enables simultaneous parallel
transactions instead of singular sequential transactions. Together these
Dual Independent Bus Architecture improvements offer up to three times the
bandwidth performance over a single bus architecture processor. In
addition, the Dual Independent Bus architecture supports the evolution of
today�s 66 MHz system bus to a 100 MHz system bus within the next year. 

The Pentium II processor and the Dual Independent Bus Architecture will be
housed in a new package technology called the Single Edge Contact (S.E.C)
cartridge. This new cartridge package and its associated "Slot 1"
infrastructure provide the headroom for future high-performance processors
and enable the broad availability of Pentium II processors. 

AGP = Improved Graphics Bandwidth Performance 

By year�s end, the Accelerated Graphics Port will make its debut in Pentium
II processor-based systems, enhancing visual computing by providing greater
memory bandwidth to the graphics subsystem. The AGP interface is a
specification for the PC platform that will enable new levels of 3D
performance and realism on mainstream PCs. AGP has achieved considerable
industry support, and initial systems will be available toward the end of
this year. AGP is an industrywide specification driven by Intel. 

Intel, the world's largest chip maker, is also a leading manufacturer of
personal computer, networking, and communications products. Additional
information is available at http://www.intel.com.
T.RTitleUserPersonal
Name
DateLines
678.1500 MHz Pentium II Demo - Real or Hype?NPSS::NEWTONThomas NewtonFri Apr 11 1997 16:349
>   ... expected to reach clock rates beyond 500 MHz by the year 2000 ...
>   demonstrations included ... real-time 3D rendering on both a 300 MHz and
>   500 MHz Pentium II processor-based workstation

So does that mean Intel has 500 MHz Pentium IIs running in their labs?

Or did the demonstrations merely simulate the expected performance of such a
beast?
678.2300 mhzROM01::OLD_CIPOLLABruno CipollaTue Apr 22 1997 10:448
    
    
    +                INTEL READIES 300MHZ PENTIUM II
    
    Intel Corp will include a 300MHz Pentium II in the announcement
    early next month, Computer Reseller News Online hears - at a
    price that will be as hot as the part: $1,980.
                                                     
678.3pricesROM01::OLD_CIPOLLABruno CipollaTue Apr 22 1997 10:4518
    +                INTEL PLANS PENTIUM PRICE CUTS
    
    Intel Corp announced Friday that all Pentium chips on the
    market by the end of 1998 would include MMX instructions.
    Separately, the Pentium II with MMX arrives on May 6, but
    according to PC Week, the company will make its cuts on the
    present its Pentium and Pentium Pro lines and Pentiums with MMX
    ahead of the news. It hears that the price of the Pentium Pro
    will barely move, falling to $415 from $428. The 200MHz Pro
    with 256KB cache will be $515, off just $10; the 166MHz and
    200MHz non-MMX parts will be cut about 30% and about 38%
    respectively, so a 200MHz processor falls to about $295, a
    166MHz to $195. Cuts on 166MHz and 200MHz MMX- enabled
    processors are seen at 25% and 8% so the 166MHz falls to $255,
    and the 200MHz at about $465. The 233MHz Pentium II are
    expected to be priced at about $590, and the 266MHz at about
    $725, and all prices are for orders of 1,000-up.
    
678.4Real!BSS::F_BLANDOJe suis grand, beau, et fort!Tue Apr 29 1997 18:145
I was at Andy's presentation in Houston.  He showed a dual processor Pentium II
300MHz workstation.  He also showed a 450MHz single processor Pentium II
workstation (The Intel press release is wrong, it was 450MHz, not 500MHz).

Both were extremmly impressive and very real.
678.5the juggernaut rolls on ...TROOA::MSCHNEIDER[email protected]Wed Apr 30 1997 01:062
    Yes we should never underestimate Intel and what they are capable of
    delivering.
678.6ironic, isn't it?CUJO::SAMPSONWed Apr 30 1997 01:112
	Yes, especially with plenty of 64-bit AlphaServers
as development platforms!
678.7COMICS::CORNEJWhat's an Architect?Wed Apr 30 1997 04:446
    re .6,
    
    This would be a nice story to leak out :-)
    
    Jc
    
678.8MOVIES::WIDDOWSONRod OpenVMS Engineering. Project RockWed Apr 30 1997 06:232
    ...Of course their Fabs all run Vax/VMS.  I always thought we should
    get `VAX inside' engraved onto each Intel chip !!!...
678.9YIELD::HARRISWed Apr 30 1997 08:466
>    Yes we should never underestimate Intel and what they are capable of
>    delivering.
    
    Is any vendor currently delivering systems with anything better than 
    200Mhz Pentium Pro chips in them?
    
678.10SUTRA::16.36.2.107::BatsSpeeding, speeding, I'm always speedingWed Apr 30 1997 09:1313

	No.

	Only 200Mhz Pentium Pro can be had.
	
	Beginning of next month you will see the Pentium II 233MHz and
	266Mhz ones. (At least from us, being PCBU)

	233MHz Pentium Pro ones (with 1MB cache) should be around June
	if I remember correctly.

	Pjotrr
678.11I have no Intel inside knowledgeCUJO::SAMPSONWed Apr 30 1997 10:113
    I should hasten to add that I don't know what platform(s) Intel
    actually uses for development.  AlphaServers would be a good
    (unofficial) rumor, though.  Intel can certainly afford the best.
678.12rumors...FORBIN::WILKINSONWed Apr 30 1997 10:5724
    Reports on the usenet suggested that some Pentium II inventory had 
    make it to Japan already and are available in retail stores.
    
    Can anyone confirm the following:
    
    Pentium II in the speed range 233MHz to 300MHz is 0.28u CMOS.
    Above 300MHz is 0.25u CMOS.
    
    The L2 cache accesses are double tick accesses for Pentium II
    as compared to single tick accesses Pentium Pro.  So, for
    a given clock speed, Pentium II has half the L2 cache bandwidth
    of the Pentium Pro.
    
    Can anyone answer the following questions:
    
    Does the Pentium II L2 cache use the same SRAM as the Pentium Pro?
    Or is it commodity SRAM in a custom package?  If proprietary, what is the
    process technology roll out for the SRAM?
    
    Hugh
    
    
    
    
678.13on caches....PERFOM::LICEA_KANEwhen it's comin' from the leftWed Apr 30 1997 13:121117
|   The L2 cache accesses are double tick accesses for Pentium II
|   as compared to single tick accesses Pentium Pro.  So, for
|   a given clock speed, Pentium II has half the L2 cache bandwidth
|   of the Pentium Pro.                               
    
    A couple of points.  Yes, the L2 cache runs at half speed.
    
    But the L1 I and L1 D cache is double the size, to 16KB I and
    16KB D.  The fastest Pentium PRO shipped ran at 200MHz, with
    the L2 cache running at the same speed.  The Pentium II at
    300MHz would have a 150MHz L2 cache.  (1.2GB/sec bandwidth.)
    
|   Does the Pentium II L2 cache use the same SRAM as the Pentium Pro?
|   Or is it commodity SRAM in a custom package?  If proprietary, what is the
|   process technology roll out for the SRAM?
    
    The cache in the Pentium Pro was a single chip packaged side by side
    with the CPU chip.  The Pentium II does not go this route, instead
    a number of different cache designs can be implemented.
    
    Anyhow, for details, here's Andy Grove's keynote from the Compaq
    hosted conference.  May is just around the corner.
    
    ( http://www.intel.com/intel/people/asg/asg_innov.htm )
    
    								-mr. bill
    
    
                    Innovate �97 Keynote Address
                        Dr. Andrew S. Grove
                           April 8, 1997
                         in Houston, Texas

 Good morning.

 My subjects are about these three points; I would like to tell you,
 very, very briefly, about Intel and what we do to bring that little
 Intel Inside� sticker on Compaq computers and, following on the
 same theme, I would like to tell you a little about what is coming
 in the future in terms of our processor road map and end on talking
 about some of the larger issues that face computing in the years to
 come.

 In its most basic essentials, Intel�s strategy can be summed up in
 one visual presentation that you see there; it�s kind of Intel on a
 post-it note. We work to advance the art of microprocessors in two
 directions; on the horizontal directions you see the increasing
 density that comes about by the finer and finer line width on
 silicon processing. In the vertical direction you see improvements
 in microprocessor architecture that, for any given technology,
 bring higher performance.

 So, the flow of chips from the upper left to the lower right is
 brought about by the combination of ever-increasing processor
 technology, ever-refined processor technology and ever-refined
 microprocessor architecture.

 On this particular chart we started with the i386� and ended with
 the Pentium� Pro processor. The story actually started 25 years ago
 with products long before the 386 that I didn�t show this on this
 chart, and in a like fashion, the story doesn�t end with this
 chart. Picking up from the Pentium Pro processor, the future has
 the same kind of advancement ahead of us. Ever refining processor
 silicon line width going substantially below the 1 micron geometry,
 to .25 micron, to .18 micron of fineness and ever refined future
 architectural generations that bring better performance for each of
 those silicon performances.

 That�s what we do for a living. That�s what we wake up in the
 morning worrying about and that�s what we go to sleep dreaming
 about. But in between waking up and going to sleep we spend a lot
 of money on this, both in terms of developmental spending -- where
 I had the opportunity to visit with a local research center
 yesterday here and somebody pointed out to me that the entire
 federal government health research budget is less that what we
 spend on R&D per year. And I don�t know which of us is wrong here,
 but something doesn�t fit with that picture. And that is the lesser
 of what we do because then what we produce with that R&D program,
 given the size of the industry, requires a very massive capital
 outlay years in advance of the need for utilization of the product.
 So this has been a chart that has grown from the multihundred
 million dollars, from the beginning of this decade, to something in
 the vicinity of $4.5 billion capital outlay budgeted for l997. And
 this is a very abstract number, even, I have to admit, even to me,
 and I cannot figure it out.

 Just flip through a few of the projects; we are under construction
 around the globe and I can�t show you everything but this is a
 major extension of our Leixlip, Ireland, wafer fabrication plant.
 We have outgrown our principal location in Penang, Malaysia, so we
 are building a new site about thirty miles away in a place called
 Kulim. We are building assembly and test facilities in China, a
 Flash memory facility in Israel; we are also expanding assembly and
 test facilities outside of Manila in the Philippines; and we have
 started a new design, engineering and manufacturing facility near
 Seattle; and having run out of all other states we even come to
 Texas. This is going to be a very major wafer fabrication plant in
 that little square or rectangle that you see there.

 So, that�s what we do. Now, where does all of this investment take
 us and what are the basic directions that we pursue in the next
 several years to come; and I would like to paint for you the
 processor road map that we are working on and that you will
 probably see before the next time that we have a opportunity to
 review progress on this.

 Basically, as suggested in an earlier slide, we are doubling our
 developmental path by doing a two-pronged development, continuing
 to expand the 32-bit microprocessor line, where you see the Pentium
 Pro processor and the Pentium II processor -- which is a product
 that we will introduce in the next few weeks -- that is where we
 are today. And the products that are labeled future 32-bit
 processors, those are all real products, with real people, with
 teams of multiple hundreds of engineers working on them as we
 speak.

 Simultaneously with those we have started a new branch which is the
 64-bit architecture of the Intel Architecture extension
 intellectual product line that is going to run 64-bit software and
 it will run all of the 32-bit software, as well. This represents a
 whole different computer architecture and the two will probably
 occupy different portions of the computing space; we intend to
 carry those two in parallel as far as we can think of.

 Now, all of this is necessitated by a variety of changes in
 computing, and two-thirds of the computing market is driven by the
 needs of business. So I think it is worthwhile for us to kind of
 step back and take a bird�s eye view of developments in business
 computing, specifically, and, for the interest of focusing on one
 area, I�m going to particularly focus on business desktop
 computers.

 The major events of the last several years have been a transition
 of the traditional business desktop that was represented by
 products like the i486� ; and products like the PCI bus structure,
 to a new generation of business computers that are represented
 today, typically built around the Pentium processor, built with the
 PCI bus structure and typically endowed with Fast Ethernet and
 audio cards and very heavily doing communications applications
 represented by electronic mail.

 We are about to make a transition to the new business desktop, in
 the years ahead of us, that I will call "the Visual Connected PC."
 This computer, typically, will be represented by a product like the
 Pentium II processor, the PCI bus structure will be superseded by a
 new bus architecture that we will call Dual Independent Bus
 architecture, and you�ll hear more about that in a few minutes.

 Computers like that will have Internet access built in. They will
 be video conferencing enabled, typically they will be connected to
 a much higher bandwidth 100Mb Ethernet, and they will come with
 built in manageability.

 Now the details are going to, of course, vary and the details will
 unfold a little differently on that but the basic characteristics
 that I want to call attention to is that the step between the
 current business desktop, the Pentium, PCI business desktop, to the
 new Visual Computing Connected Visual Computer, based on the
 Pentium II processor and the Dual Independent Bus structure, is as
 large a step in its combination as the step from 486 and
 ISA(phonetic) to Pentium and PCI was. There�s a very major
 discontinuity and this step and this evolution brings about a
 redefinition, a necessary redefinition, of how we look at processor
 performance.

 Processor performance used to be looked at as a one-dimensional
 phenomenon; it is now necessary to use three different vectors to
 describe performance; integer performance, floating point
 performance and multimedia performance, because all of these are
 encompassed in the Visual Connected Computer. These require
 different technical attributes to achieve.

 In the case of integer performance, the architectural goodies that
 we put into the P6 architecture -- whose first instantiation was
 the Pentium Pro processor, the second one is the Pentium II
 processor -- that we call dynamic execution, predominantly gives
 you highest integer performance, highest CPU core performance.

 The cache bus that, again, you will hear in a moment, I�ll just
 forward reference it, that is a key part of the Dual Independent
 Bus architecture will allow the extension of floating point
 performance by allowing floating point, data that requires floating
 point calculation, to bypass the ordinary bus structure and
 communicate with the floating point unit and those parts of the
 subsystem that requires independently, and, lastly, MMX� ;
 technology that we introduced earlier this year, leads to faster
 communication and multimedia applications.

 I would like to demonstrate what these things look like and what
 these things will look like in the future, so let me first ask
 Spencer Clark from Intel to show us a demonstration of ...

 Spencer Clark: This is a dual Pentium II processor system running
 at 300MHz and what I�m going to show you today is an application by
 a company by the name of SolidWorks and what I have here is ...

 Dr. Grove: SolidWorks is a CAD application, right?

 S. Clark: That�s correct, that�s a CAD application, this particular
 object here is a very complex object; it�s a SCUBA regulator and
 I�m going to go ahead and rotate the object. What we�re looking at
 is an object that has a lot of lighting, a lot of shading
 characteristics to it, very many polygons required to render this
 as I move the object.

 Dr. Grove: So, every time that object moves, every point and every
 bit of, every piece of data that is representing that point and the
 lighting on that point, gets recalculated over and over again in
 real time.

 S. Clark: That�s correct, it�s very compute intensive. What I am
 going to do now is I�m going to show you a subpart of this
 particular model, which is the mouthpiece, bring that up, and
 you�ll notice that I can take this particular subpart and move it
 very smoothly. It actually flows very nicely.

 Dr. Grove: Presumably you can move it through there because there
 is less, fewer data elements involved in it, so therefore the
 computer, at the same data rate, is capable of recalculating it
 more often per each second.

 S. Clark: That�s correct, it�s a less complex module than what we
 had up there before. OK, what I would like to also do is show how
 we can do, very quickly, on-the-fly manipulations to this
 particular object.

 What I am going to do is, let me get in a little better position
 here, we�re going to apply afillet to this particular part of the
 object. If you look at it at a very sharp right angle here, we want
 to smooth that out, I�m going to do my best to see if we can just
 grab that, one second here, this is the tricky part getting that
 one part that we want, and that�s it right there.

 Dr. Grove: Well, don�t let these guys make you nervous.

 S. Clark: OK, easy for you to say, I guess. We�re going to, I�m
 going to go ahead and apply a fillet here of 15mm and that will
 make the transition at that particular part of the object a little
 bit smoother. And you can see, we did that on-the-fly, and it no
 longer has that real sharp angle there.

 Dr. Grove: So the mechanical calculation immediately translates to
 its visual representation with a click of a button.

 S. Clark: That�s correct. Now we have this and are just showing
 this in an engineering sense but we can also bring up this
 particular application within an office product, Microsoft* Word*,
 for example.

 Dr. Grove: Which is a key point because unlike many workstations,
 this, when all is said and done, is a personal computer and it runs
 PC applications like Office* or like Word.

 S. Clark: Exactly, and we�re actually using (unclear) here. What
 I�m going to do is I�m just going to type something in this to show
 that it�s just a standard Word document. Well, as you can see,
 we�re just adding text to it but what�s nice about this, typically,
 with office products we started out with just the ability to have
 text and then later we transitioned into having text and graphics;
 well, now, we actually have an interactive 3D object within the
 office.

 Dr. Grove: So, for instance, an engineering memo will come with the
 text, with the explanation, with the mechanical part, built in as
 an object in that demo; subject to the reader�s whim, he can rotate
 it, he can look at it and this could be an engineering document, it
 could be a medical document, it could be anything that involved a
 3D object, all will be part of the document of the future.

 S. Clark: That�s correct. I just double-clicked on this object and
 we now have the toolbars here on top, I can now go in and do what
 we did before when we were in the SolidWorks application and I can
 rotate this object.

 Dr. Grove: But you�re inside Word now.

 S. Clark: That�s correct.

 Dr. Grove: Very good. Thank you.

 Progress does not stop with a dual 300MHz Pentium II processor and
 we�re going to push the envelope a little bit and pushing the
 envelope is Steve Hunt from Intel who will demonstrate

 Steve Hunt: Hi, Andy. We�ll be demonstrating a technology
 demonstration of the Pentium II microprocessor and what I want to
 do is, kind of, give you a little, quick tour.

 You can see the actual Pentium II microprocessor, here, the silicon
 and the package around it. You referred to the cache bus, and this
 is a very important element of the Pentium II processor shown here
 on this part of the I/O and this interfaces to a variety of
 different types of L2 caches. Here, we have a separate system bus
 that goes off and talks to the DRAM and the I/O.

 Dr. Grove: More about all of that in a few minutes.

 Steve Hunt: This is, kind of, the die and you can see we�ve
 actually doubled the on-chip primary cache. Up here, we see the
 floating point unit. Now this is a very powerful floating point
 unit and what we�re going to do is run an application where we�ll
 actually try to stress that floating point and show how the
 floating point execution bandwidth scales along with the processor
 clock frequencies. Now we�re running, right now, at 450MHz, I guess
 I�ve got a rounding error here, 449, but ...

 Dr. Grove: Don�t exaggerate.

 Steve Hunt: OK, 449, and what we�ll do, let me go on ahead and fire
 up the, an application from Caligari called trueSpace, it�s a 3D
 rendering package, what we�ve done here is downloaded a model off
 of the Internet, this is a standard benchmark that is supplied by
 the company and we�ll actually go ahead and kick it off.

 This particular model renders on a 200 MHz Pentium Pro processor
 with 512k cache in 24 seconds, we�re rendering this in about 12-13
 seconds, roughly at 2x speedup. So this is a lot of complex
 lighting calculations and texture maps being applied here.

 Dr. Grove: And that speed is a testimonial to the higher speed
 floating point unit but also to the improved bus structure that
 allows that floating point unit to do its thing at a system level.

 Steve Hunt: Exactly. We�ve got to feed that raw floating point
 engine with the op RAMs and instructions associated with this
 program.

 Dr. Grove: Do you have any livelier demo than that one?

 Steve Hunt: Sure, let�s see what we can find here, I�ve got one
 that will fire up, it�s a, kind of a more interactive, 3D
 environment. Everything is completely done in real time. Now this
 is a classic division of responsibility between the processor and a
 3D graphics card; we have the 3Dfx coupling with the Pentium II
 processor. In this case we have the Pentium II processor doing all
 of the geometry calculations and the pixel seal, in other words,
 sealing all the polygons that are created by the Pentium II is
 accomplished by the 3Dfx card.

 Dr. Grove: It�s sort of like a coloring notebook, where the Pentium
 II processor does the geometry and the 3Dfx card does the coloring.

 Steve Hunt: Exactly. So it does the rendering. OK, so we�re running
 now, I�m going to send it on kind of an automatic fly-through for a
 moment. A couple of things I�d like to point out: You can see
 there�s quite a few objects, we�re flying through a predefined path
 at the moment, we have spatial audio so objects in the scene that
 have sound associated with them become louder or softer as you
 become closer to them, the function of your position in the scene.
 We�re going to fly by this barn and I�m going to put it into kind
 of an interactive mode because I run the farm now. I want to point
 out a really neat feature here in this application which is
 reflection mapping. So, you can see the reflection of ...

 Dr. Grove: You can see the clouds in that pond.

 Steve Hunt: That�s a lot of, you know, complex ray tracing types of
 algorithms, so we�ll take it on around here. I�m actually flying it
 interactively now.

 Dr. Grove: What is that video-looking thing there?

 Steve Hunt: Well, let�s go and take a closer look. I think I hear a
 familiar theme here ...(music)

 Dr. Grove: Not again ...

 Steve Hunt: I think it�s the Bunnysuit, let�s see, we saw that
 during some famous event. See, I�m kind of on the front end of the
 blimp and the sound goes away a little bit; let me take it in a
 little closer, I don�t want to hit that helicopter ...

 Dr. Grove: I think we�ve had enough of that one.

 Steve Hunt: Oh, OK, you want me to, I can go on ahead and ...
 (crashing sound)

 Dr. Grover: Thank you, Steve.

 The point I�m trying to make and the point that we try to bring to
 you through these demonstrations is that it is no longer enough to
 deliver faster CPUs; we must deal with and cross the bandwidth
 valleys of death where, on the one side, we can crank up CPU power
 but unless we cross that valley of death we don�t come out on the
 other end with that same performance.

 And I use the term in plural because there are two elements to
 that: There is the bandwidth that we have to deal with between
 processor and memory, and the bandwidth that we have to deal with
 between processor and graphics subsystems.

 We were very, very proud, five years or so ago, when, in the first
 major revamping of PC architecture, our industry delivered the
 current architecture, bus architecture. I described this as the big
 step from ISOP 486 to Pentium and PCI bus and it was a very major
 improvement at the time. But, as it turns out, with today�s
 processor speeds, this five-year-old bus structure is becoming an
 increasing limitation.

 Limitation comes in two places. One is increasingly relying on L2
 cache, as Steve pointed out to you in the floating point
 calculations, being a perfect example of it, and the access to this
 L2 cache is limited, and the current technology, is limited to
 66MHz -- way lower than the clock frequency of the microprocessor.

 Similarly, the access to the graphics subsystem has to go through
 the, system bus has to go through the PCI bus and is limited there.
 And with the increasing demands on the system coming from 3D
 graphics, coming from 100Mb Ethernet, coming from external
 bandwidth delivered through technologies like 1394; this limitation
 will increasingly limit us. These bus limitations will increasingly
 limit system performance.

 Simply put, the three vectors of performance demand a different bus
 architecture, and the bus architecture that we are ushering in this
 year is what we call the Dual Independent Bus architecture.

 In its simplest form, in this bus architecture, we break the bus,
 the memory bus, into two parts. One addressing the L2 cache and one
 addressing the system. The most important of those, again, as Steve
 was trying to steal my thunder a minute ago, is the bus that goes
 to the L2 bus that, for simplicity, we�ll call the cache bus. This
 has the largest amount of data traffic, in many instances 95
 percent of application performance is traversing this particular
 bus.

 The beauty of the microprocessor architecture that is implicit in
 the Pentium Pro processor and the Pentium II processor is that it
 contains a technique called out-of-order execution that can manage
 the data stream in advance and split them up between these two
 buses and use them very judiciously.

 So let me talk a little bit about these two buses. The first of
 those, again, is the cache bus, this is the traffic between the
 core of the central processing unit and the L2 cache, as I
 mentioned, 95 percent of application bandwidth is taken up by this
 and, specifically, in the case of floating point applications, as
 you saw in Steve�s earlier demo, it is very heavily dependent on
 this cache bus.

 The point here is now that with this bus architecture we are no
 longer limited by the 66MHz system bus for this traffic, but the
 power of the bandwidth of this connection scheme will scale with
 the processor, so, as the processor speeds up, the bus will be able
 to keep up with it and not provide a limitation.

 The second bus, the system bus, likewise, first of all, no longer
 carries that traffic and, second, it is scaleable itself. Year by
 year, we will be able to speed this thing up, keeping up with the
 generations of microprocessors and the portion of the traffic that
 is directed to the system bus and the key here is to deliver this
 to the mass market, to the mass desktop market, in a cost effective
 fashion. And we will do that in a fashion that we call the "Slot 1"
 implementation which, again, will come up in a few minutes.

 Backing away from the details of this, what we are doing with the
 Dual Independent Bus architecture is addressing the total processor
 bandwidth by adding up the bandwidths coming from the cache bus to
 the bandwidths coming from the system bus. The total system
 bandwidth is the sum of these two and to give you an illustration
 of 300MHz Pentium II processor that you saw before, the cache bus
 will be operating at half the processor, in this case, half the
 processor frequency of 150MHz, corresponding to 1.2Gbytes/sec of
 bandwidth; we add to that the system bus bandwidth of 66MHz at
 5.28Mbytes/sec. When we total these up, the total processor
 bandwidth is going to be 1.7Gbytes/sec capability. Over threefold
 of the bandwidth of today�s Pentium processor with PCI
 implementation, a very major increase in total bandwidth and, in
 addition to that, scaleable with processor performance, so with
 headroom for the future.

 This bus architecture was basically first pioneered with the
 Pentium Pro processor and was hidden in a complex package that was
 fairly expensive and not designed with the idea that this bus
 structure will spread through the entire range of PC performance.

 What I mentioned earlier with the phrase of Slot 1 is an
 implementation of the same thing with an eye toward mass
 production, with the ability to use industry standard DRAMs and,
 due to the connection scheme, a processor bandwidth, a system
 bandwidth, that will further scale with increasing frequencies.

 This Slot 1 implementation of the Pentium II processor looks like
 this, it is a little cartridge that has an H-connector that plugs
 into the system and it is designed from the ground up with mass
 production and low cost, cost effectiveness, in mind. And, in fact,
 as we are ramping this into production, we are finding just one
 thing that is really giving us headaches; it has a little
 holographic version of the Pentium II processor chip on it and
 attaching that to the cartridge is turning out to be quite a
 difficult production task, but we will overcome that one too.

 Let me move on to the second bandwidth, second valley of death,
 bandwidth valley of death, which is the connection between
 processor and graphics subsystems. What we are talking about here
 is, on your left of the system wherein the Dual Independent Bus
 architecture is using what we call the Accelerated Graphics Port or
 AGP, we can shunt off the data stream to the graphics subsystem
 independently of the cache traffic and the memory traffic and
 therefore deliver high and scaleable bandwidth to the graphics
 subsystem.

 This capability is again related to the capability of the Pentium
 II processor which has inherent parallelism built into it that is
 capable of managing the data stream in this type of fashion.

 Scaleability of this bus is just as important as the scaleability
 of the cache and the memory bus because graphics subsystems will
 become faster along with microprocessors and we need to make sure
 that we are not building in a future bottleneck.

 And, in fact, the Accelerated Graphics Port bandwidth will scale,
 its �97 implementation is at 133Mbytes/sec, when you look at it our
 plans, and these are supported by solid technical work, imply a
 doubling of AGP bus bandwidth every year all the way through the
 end of the decade, which, actually, interestingly enough, is a
 little faster than Moore�s Law predicts transistor and chip
 performance which doubles only every 18 months.

 So, I think we are getting ahead of microprocessors with the
 graphics bus system for the first time in the history of personal
 computing. The point here is the way we are getting across these
 bandwidth valleys of death, we engineer our way through. The memory
 bandwidth will be delivered with the Dual Independent Bus
 architecture, the graphic bandwidth will be delivered through the
 Accelerated Graphics Port working in conjunction with the Dual
 Independent Bus architecture. All of this has the benefit, very,
 very simple benefit, all of this pretty complex technology, will
 enable the computer to scale with processor performance and deliver
 that performance in a software visible, a user-visible fashion, all
 the way to the use.

 So, that�s kind of our story on processor and platform roadmap and
 with that I would like to move on to my last subject which is some
 of the trends which face us, computer users and computer
 manufacturers, all of us in the computer industry in the years to
 come.

 I would like to start by submitting to you a basic premise for the
 future: All computing is going to be network computing. This is
 almost a given in business computing, but actually it is going to
 be a given in consumer computing and one of the things as kind of a
 subtext that I would like to leave you with is there�s going to be
 an increasing blurring between consumer and business applications
 in the very near future, but more about that in a minute.

 The first megatrend that we are dealing with in this network
 computing area is the mandate to reduce the total cost of ownership
 of network computers; and the way to achieve this is like the way
 we engineer ourselves through any other problem, is to build this
 into every machine, right as it is manufactured, and we can do that
 and we must do that. We can and we must instrument every personal
 computer that we ship out of the factory.

 Once we instrument and manage a PC, this management will allow for
 a network administrator in a central location to obtain system
 information in a standard fashion, enable remote diagnostics and
 repair and we will be able to do so without giving up the
 flexibility and adaptability of personal computers that has been so
 successful and so powerful in making personal computers successful
 in the last 10 years. The answer to this management problem and the
 answer to the total cost of ownership problem is a spectrum of
 managed PCs.

 What do I mean by a spectrum? The spectrum comes first in the
 dimension of various degrees of flexibility, ranging, on your
 right, the fully flexible, adaptable PC as we know it, full range
 of applications with flexible configurations, user upgradeability
 and remote management built-in, as I said before.

 On the left side we can see a version of the personal computer and
 remove some of the flexibility, orient it to applications and users
 with more narrow range of applications, limit the configurations,
 limit, or eliminate, user upgradeability altogether, and yet
 deliver the same type of remote management capability across the
 entire spectrum of flexibility in this horizontal fashion.

 The second axis of the spectrum that we are talking about is both
 flexible and Net PC capabilities and constructs can be delivered in
 a range of performances dependent on the present and anticipated
 user needs just as it has been the case with personal computers.
 Let me proceed to a demonstration and welcome Jeff Lo here, who is
 a new employee of the Data Integrity group. Welcome aboard, Jeff.

 Jeff Lo: Thank you, Andy.

 Dr. Grove: And what you have here is the first ever public showing
 of, will I get electrocuted if I touch this? This is a Net PC,
 hopefully alive and well and fully managed, and a new employee gets
 a brand new Net PC on his desk on his first day of work and he
 checks in, proceed.

 Jeff Lo: Well, I have been playing around with this Net PC a little
 bit and checking it out but I think I need to go check out the
 company orientation, so I�ll go and click on the
 "Welcome" button and we�ll see our person here welcoming us and I
 think she�s supposed to be telling us something but I�m not hearing
 her say anything.

 There seems to be some sort of a problem here, oh, the LANDesk�
 Client Manager has detected a problem, it says I can click on the
 "Solve" button. Well, I�ll let it try and fix this itself, so I�ll
 click on "Solve" and it has already figured out it is missing some
 software and has already gone ahead and installed it ...

 Dr. Grove: Now, you didn�t do this.

 Jeff Lo: I didn�t do this, the Client Manager just ...

 Dr. Grove: System in the clouds did it for you.

 Jeff Lo: Yes, uh, huh.

 Dr. Grove: That�s the kind of computer I�d like, too.

 Jeff Lo: There�s still another problem but it doesn�t know how to
 fix this one so I can click here to contact support. Let me go
 ahead and do that and it has already entered my name, my e-mail
 address and what it thinks the problem is and let me go ahead and
 click on "Submit" here and see what happens.

 Matt Arnet: OK, the ticket has been submitted with the unknown
 audio problem. This took hours at my old company, somebody always
 had to come out and do this, see what happens here, OK, in the
 not-so-distant past the network administrator would have to receive
 a phone call or e-mail and physically get himself up and go over to
 the client and fix the problem and obviously that would increase
 your total cost of ownership. Now these network administrators are,
 typically, called Help Desk technicians and they would be
 monitoring a screen such as what I have here on my LAN Desk Client
 Manager. Here you see that throughout my busy day I have closed
 quite a few tickets here of previous problems or occurrences that
 have transpired on the network.

 Now, if I just refresh, I�ll see that a Mr. Jeff Lo has had an
 unknown audio problem, L2 critical, and this is open, OK, well
 that�s a new ticket so, what I�m going to do is I�m going to look
 into this a little more deeply.

 So, I�m going to double click on Jeff Lo�s ticket, bringing up some
 more detailed information, again, you see that the employee, Jeff
 Lo, unknown audio problem, and this was all discerned on his
 computer via the LANDesk Client Manager. And it was shipped over to
 our inventory database.

 Device data, Creative Labs Sound Blaster*, OK, so that tells me he
 does have this equipment in his Net PC, so that�s probably not the
 source of his audio problem. Time stamp is that it occurred today
 and less that five minutes ago.

 The next LANDesk Client Manager recommends that I take remote
 control of his PC and see if I can rectify the problem over our
 corporate Internet. OK, I�m going to take control over TCPIP,
 manage his PC and take control of it just as if I was sitting
 there. So, I�m going to give Jeff a call here ... (ringing)

 Jeff Lo: Hello, this is Jeff.

 Matt Arnet: Yes, this is Matt Arnet over at the Help Desk.

 Jeff Lo: Oh, great, that was quick.

 Matt Arnet: Well, yes, I recognized that you had submitted a ticket
 here and I wanted to see if I could help you by taking over
 remotely and see what the source of the problem may be.

 Jeff Lo: Great, if you can fix it, that would be great.

 Matt Arnet: OK, let�s see, right now, I should be accessing through
 our pipe, there we go ...

 Jeff Lo: Wow, my mouse is moving all by itself ...

 Matt Arnet: Well, yeah, yeah, don�t do anything there because I�m
 in control of your PC at this point, now, you haven�t been playing
 with your computer at all because I know you�re a new hire and you
 did go through the new employee orientation, did you not?

 Jeff Lo: Yeah, but it was this new Net PC, I wanted to check it
 out, see what it could do.

 Matt Arnet: Now, didn�t they tell you that before you start playing
 with your PC you had to have additional training before you changed
 any of the configuration parameters?

 Jeff Lo: I know, I know, but I thought ...

 Matt Arnet: OK, well, you know, I�m going to go ahead and fix it
 for you this time but I�m going to make a special note that we do
 not accept any more tickets until you have been authorized to
 submit, so what I do see down here in the lower right hand corner
 is that his microphone, or his speaker, has been nulled. What I�m
 going to do while I�m in control of his computer, and this is
 happening on his computer, is I�m going to bring that up, because I
 have a feeling that that�s the source of his problem.

 Now, over here, low and behold, he has been muted, so I�m going to
 uncheck his mute, close that master output window and I�m going to
 go back and tell Jeff that I�m going to refresh this, thus taking
 him back to the initial screen and Jeff, I�m going to go ahead and
 close out because I have a feeling that I fixed your problem for
 you so why don�t you go ahead and launch your application that you
 tried before.

 Jeff Lo: OK, I�ll try it again here.

 (Woman�s Voice) Hello, I�m Allison ...

 Jeff Lo: Oh, great, great, that was no sweat at all and it�s
 working, thanks a lot, Matt.

 Matt Arnet: OK, thank you, Jeff, and don�t call us, OK, we�ll call
 you.

 Dr. Grove: Thank you.

 What you have seen courtesy of Jeff Lo and Matt Arnet is two new
 things, working together; one of them is the Net PC, that I pointed
 out, the second one is the management task, the remote management
 task that was executed was done on the basis of an Internet-based
 management scheme called Web Based Management which is an
 initiative that we are working with Compaq and Microsoft to bring
 to the market to use the pervasive deployment of Internet based
 networks to be the foundation for all management software, remote
 management software. We have at least gotten it to the point where
 we can demonstrate it.

 So, that�s about the spectrum of managed PCs; both managed and the
 Net PC being a key element of what we are talking about.

 The second megatrend is the emergence of the Java* programming
 language as a tool for unlocking legacy data, a large portion of
 which continues to reside in legacy computers, in legacy databases
 and is very important for us to capture those and integrate them
 with today�s network computing environment. Java is a very
 important tool in this one and Intel�s role with respect to Java is
 to make it run the best, meaning the fastest, on Intel Architecture
 machines.

 So, I would like to give you a bit of a status demonstration of how
 we are doing on this one and what I want to do here is demonstrate
 it by a bake-off, basically, what we have is the same Net PC that
 you have seen before, that is on your right screen, the data
 pertaining to the Net PC is on your right screen.

 Here, we have Sun JavaStation operating with a microSPARC and the
 data pertaining to that is on the left screen. So it is a little
 confusing this way but keep in mind Net PC, JavaStation. And what
 we have in the JavaStation is 100MHz microSPARC memory, no hard
 drive, hard Java browser and all the rest of it is the same in both
 instances. In the case of the Net PC, as we saw earlier, we have a
 233MHz Pentium II processor, same amount of memory and, of course,
 it has a hard drive, Windows NT* workstation and Internet
 Explorer*. So, let us see this demo, Spencer and Jeff will do for
 us and the first part of this demo, Jeff, walk us through this,
 please.

 Jeff Lo: Well, the first thing that we�re going to do, is we�re
 going to load the caffeine mark, benchmark and this is a benchmark
 of Java code, bring it up into the window here and then we�ll go
 ahead and start it and I�ll tell you what is going on, ready, go.

 The way this works is that it runs a bunch of Java code in a loop
 and it counts the number of times it goes through a loop in a
 specific amount of time and displays that number. So, bigger
 numbers are better, we�ve gone through it faster, more times
 through the loop, takes about a minute for the whole thing. This
 first test you see the score. I�ve got 257 over here on the Net PC,
 and Spencer, on the Java, is 113. The next one, the loop score, is
 207.

 Spencer: I�ve got l24.

 Jeff Lo: Outrunning it quite a bit. Here, we�re starting to look at
 some of the graphics benchmarks that come up, watch the speed of
 the repaints. And just some other typical things, drawing of
 dialogue boxes that you would see on your screen, and we�re done
 here, our caffeine mark score for the Net PC is 356, compared with
 ...

 Spencer: 126, on the JavaStation.

 Dr. Grove: An almost 3 to 1 difference and it is not a fair
 comparison because this is a faster processor, but if I take my
 slide rule out, and scale down the performance that Jeff got here,
 the 356, to the equivalent that we would get, assuming linear
 scaling, at 100MHz to match the microSPARC that�s run here, we
 would still get 150, for the scaled down Pentium II processor, as
 compared to 120-something for the JavaStation, so, we are still
 beating, even on this basic caffeine mark, the Java performance.
 But that�s probably not all.

 Jeff Lo: Yeah, we can look at some actual applications, not just
 benchmarks. We�ll go back here to our start page and have the
 Corell Office for Java* they�re working on and we�ll go ahead and
 start this up now. Go ahead on three: one, two, three. And it will
 take about eight seconds or so on the Net PC to bring up this
 shell, it�s already here, and it�s about 20 or so on the
 JavaStation, and there, now we�re up.

 Dr. Grove: About the same kind of scaling that we had at caffeine
 mark.

 Jeff Lo: Yeah, but we actually see even more on some of these
 applications; if we look at the calendaring software they have
 here, Infotools, we�ll start that up: ready, go, here, it takes
 only about nine seconds, it sits here for a few seconds then it
 just suddenly it will pop up. And, there we go, we�re ready, it�s
 about 35 seconds there.

 Dr. Grove: We�re not going to wait for that.

 Jeff Lo: No, I wanted to show you something else we can do here
 that you can�t do there. A big advantage of the Net PC is, it is a
 regular PC and I can go ahead and bring up Microsoft Office and
 launch Excel* or any other PC application that I�ve got on this and
 we can�t do it there.

 Dr. Grove: And that, of course, is operating at full speed because
 it is a native application running on a Net PC which, after all, is
 Intel Architecture-based PC, runs all the Intel Architecture-based
 applications.

 Jeff Lo: Yup.

 Dr. Grove: Good, thank you very much.

 If I summed these up in the raw fashion unadjusted for frequency
 you saw the caffeine marks almost 3 to 1 difference; you saw the
 difference in Corell Office; launching in the personal information
 manager application and, of course, the fact that the Net PC runs
 PC applications much the same way as that workstation, the 300MHz
 workstation ran Microsoft Word also.

 The point here is the Net PC delivers performance on Java
 absolutely as good or better than the equivalent and runs all of
 the PC applications which help us make the PC and the Net PC, the
 Intel Architecture altogether, as the no-risk, no-compromise choice
 for Java applications. No need for either of those because we can
 match the performance and you don�t have to compromise because we
 carry all the PC applications, as well as acquire all the
 Java-based applications.

 Lastly, the third megatrend is the emergence of the Internet in
 business-to-business and business-to-consumer communications.
 Excuse me for just a minute.

 The key element here is this intersection between business and
 business and business and consumer is enhanced by the prevalence
 and easy accessibility of visual computing. In an article a couple
 of months ago, John Reed was holding forth on the subject of, John
 Reed, CEO of Citibank, was holding forth on the subject of
 electronic banking and he said, "The limiting element in the
 availability, or the widespread use of electronic banking, is
 establishing trust between the bank customer and the bank."

 Nothing helps establishing that trust as much as the visual element
 that modern computing can bring to that. And, again, I think, the
 best way to demonstrate how this could work is to conjure up a
 scenario where customer to database and customer to other
 individual interfaces can be seen and enhanced by visual computing.

 And here�s Jeff again, what happened, the last time I saw you in
 Data Integrity?

 Jeff Lo: Got a new job, well, I think my true calling was sales, so
 I joined a company to do real estate sales and I understand you�re
 trying to sell your house.

 Dr. Grove: Absolutely, and, of course, nothing would please me more
 that to deal with a real estate agent that got fired for playing
 with his computer.

 Jeff Lo: Yeah, now I�ve got a new computer, too, I�ve got this nice
 laptop notebook computer, so I can go out to the customer and sell
 their houses. So what I�ve done is, before I got out to your house,
 I entered the information on your house into our database, but I
 wanted to get a picture, so before I came in I took a picture of
 the house. But, I thought, to show the prospective buyers of your
 house and the people you�re going to be trying to buy your house
 from, I�d get a picture of you that we�ll add to our database, as
 well, with my new digital camera.

 Now, what I need to do to put this on our network here is to dial
 in over the phone line in your house back to our corporate server
 which is actually sitting right here. We�re dialing up over this
 regular phone line; and connecting up, while that�s connecting I�ve
 got the picture I just took of you plus the one of the house on
 this flashcard which I can take out of the camera and insert right
 into the notebook, so that we can access it online.

 And I�ll go ahead and pull up our page here, this is our page where
 I can log in, give it our password and I�ll click "Submit" and it�s
 going to give me a bunch of options that I can go through here and
 what I�m going to look for is the option to edit the listing so I
 can add the photo of your house in. OK, here it is, "Edit
 Listings," and now it�s going to do a query on the database; all
 these pages are actually coming out of an Informix database and it
 is doing a query for listings that are mine, that are listed by me.

 Dr. Grove: Pause for a moment, all of these pictures of the
 different houses come out of the server database that resides on
 the server in your head office and you�re downloading them and
 displaying them on this computer.

 Jeff Lo: Yes. And I�ve been busy, I�ve already got six listings and
 here�s yours without the photo there. Let me go ahead and click on
 that so we can edit it and put the photo of the house in and get it
 up to our server.

 What I�m going to do is I�m going to run our FTP application. But
 let�s take a quick look at these photos that I�ve just put in here.
 To figure out which one it is I�ll open the drive that is our
 flashcard and there�s the house and here�s that picture of you up
 here on stage. Looks pretty good, I think people will sell you
 their house. But this is really the one we want, this number 56,
 so, go ahead and close this down and I�ll connect into our server.
 Give it the password for our special upload area and we�re
 connected up and I�ll go ahead and send the file now. I just don�t
 have any luck with computers, everywhere I go something happens, it
 says access is denied but ...

 Dr. Grove: Old Matt is probably standing by someplace to fix it up
 if you somehow can get to it.

 Jeff Lo: Probably, so let me do this, we�ve got our Help Desk page
 here, and I can submit another ticket here, but since it�s kind of
 urgent, I�m sitting here with you and I want to get this listing
 uploaded, I want to give him a call but ...

 Dr. Grove: You only have one line in this house.

 Jeff Lo: Right. And I�m using that to dial in to our server. So,
 what I can do is click on here and it will make an Internet phone
 call into our Help Desk ...

 Dr. Grove: On the same line?

 Jeff Lo: On the very same line.

 Dr. Grove: Wow.

 Jeff Lo: All running over Internet protocols. And it�s placing the
 call.

 Matt Arnet: Oh, oh. Jeffrey, where do I know that name from? Hello,
 this is Matt at the Help Desk.

 Jeff Lo: Oh, look, it�s Matt, you�ve gone to the real estate world,
 as well.

 Matt Arnet: Yes, Jeff Lo, you�re not in the building, are you? Can
 I help you, do you have any problems, today, or what can I do for
 you?

 Jeff Lo: Yeah, I�ve been trying to upload this image, I�m trying to
 sell Andy�s house and put his photo on line but it said access was
 denied, can you help me with this?

 Matt Arnet: OK, well this time it sounds like it�s not your fault
 so let me take a look at our server. I�m going to take remote
 control of our server from the same web page that I did before.
 It�s just the server management instead of a client management. Now
 I�m in control of our server over the Internet and I�m going to
 take a look at, since he�s dialed in via RAS, the Internet service.
 Now, Jeff, where were you putting those files?

 Jeff Lo: It�s in the new directory for the photos.

 Matt Arnet: OK, well, let�s take a look at that and make sure that
 that�s all set up for you properly, so, again, taking control of
 everything, I do see here that on the server right is not checked
 for Jeff. So I�m going to go ahead and check right, select OK,
 select OK here and, Jeff, why don�t you give that a try again
 because you should be taken care of.

 Jeff Lo: OK, I�ve got the file selected and I�m telling it to
 upload and. ...Wow, great, it�s copying it up, 60 percent 70
 percent, all set, thanks a lot, man.

 Matt Arnet: My pleasure and give us a call soon.

 Jeff Lo: I�ll try not to. Let me get back to here, so, we�ve got
 the path name of the new file up there and we�ll go ahead and tell
 it update along with all the information on the house and it�s
 going to now insert this record into the database with this new
 image. So, it�s been updated and I�ll tell it to go ahead and
 display your house. And it�s downloading the picture we just put
 up. Now, I understand that you really like the house that you�re
 in.

 Dr. Grove: I particularly like the roofline.

 Jeff Arnet: Well, what we can do is we can tell this database to
 ....

 Dr. Grove: But it�s too small.

 Jeff Lo: Too small? Too small but you like the look, so , we�ll
 tell it to find a similar house. We�ll bring up our search page and
 we�ll need what, three bedrooms or so?

 Dr. Grove: Uh, huh.

 Jeff Lo: So, three bedrooms, two baths and we�ll also tell it here
 to match the appearance so it will actually use the Informix
 DataBlade on the database to do a search on the images of all the
 houses in the database.

 Dr. Grove: It is a visual database, so it is capable of comparing
 like images by searching the database by visual criteria.

 Jeff Lo: Correct. So we�ll go ahead and click on "Search,"
 submitting that sequel request and it�s comparing the images, the
 price, the number of rooms and it�s pulling up what it thinks the
 four best choices are here. Give it a second here to download.
 Well, we can see these all have this little peak in the roof like
 your house does. It�s grabbing these last two.

 Again, a number of the same visual elements; this one also has the
 same kind of railing in front of the porch as does this one. You
 know, I went out and looked at this house the other day; it looks
 really nice, I think this is your next dream house ...

 Dr. Grove: Well, it is particularly my dream house because it�s
 only one block away from Intel so I never have to take my car and
 look for a parking space again.

 Jeff Lo: Yup, and it�s my dream house, too, because it�s the most
 expensive.

 Dr. Grove: Well, you have to keep a job for that, Jeff.

 Thank you. What we have seen here is several technologies working
 together. First of all, you have seen a consumer application where
 with the ordinary telephone line we have still used the techniques
 of Web Based Management for a service person, support person, to
 remotely fix a problem that Jeff is prone to be creating.

 Secondly, you have seen networked computing at work where a very
 data-intensive database application involving visual databases was
 used on a powerful server and yet the very data-rich image was
 displayed and manipulated on the client that�s connected to the
 computer.

 In the future I believe that commerce will be conducted
 increasingly screen to screen. In this case the screen-to-screen
 transaction took place between Jeff and Matt, the support person,
 but if it was a real real estate situation the next transaction
 could very easily have been to a mortgage broker, again, screen to
 screen, with a single phone line using the capabilities of the
 Internet to display video images telephonically in addition to
 data.

 After all, the video images and the sounds are just so many more
 bits traversing, intermixed with the numerical data that are
 involved in this transaction. Or, conversely, or beyond that, it
 could have been very easily a transaction could have transacted,
 could have taken place between seller, us, and a buyer on another
 line, again, screen to screen, using the capabilities of connected
 computing.

 Increasingly, whether it is, we are not going to jump to this
 complex situation in one move but, screen-to-screen commerce is
 coming our way very, very rapidly.

 So, if I sum up the megatrends that are facing us, all computing is
 heading, if it isn�t already, toward network computing;
 increasingly it will be visual in nature. Network computing and
 visual computing require performance. Performance, in turn, will be
 determined as much by the internal bandwidths of the computer as by
 the processing power, and what I tried to illustrate to you is how
 we are working to deliver both the processing power and the
 internal bandwidth for the future of network computing.

 Now, we are one company in a network of companies in a different
 sense of the word, that working are making network computing a
 reality and we are very, very pleased to have the partnership of an
 outstanding company like Compaq to help us, guide us, collaborate
 with us, deploy our technology and move us along when we stray.

 And, in commemorating all of that and perhaps one of the key
 elements, key successful elements, of our joint work, may I have
 Eckhard Pfeiffer join me for a moment, please.

 Recently, Compaq reached a very important milestone and Intel was
 very, very proud of them. Compaq, through their outstanding
 engineering and marketing effort, has managed to bring Intel
 Architecture-based servers into places where, until a few years
 ago, we never dreamed it would be possible; and are turning their
 work with, hopefully, a little contribution from us, are turning
 high-performance servers into a volume business.

 In commemorating the one millionth server, here is a framed ad of
 ours that we ran to congratulate you with that.

 Mr. Pfeiffer: Thank you, Andy. I�m deeply touched. Frankly, I
 didn�t expect it. You are the one to be thanked and receive a
 little gift, but before handing it over to you, let me thank you
 for this outstanding presentation, for this fabulous demo.

 I think you have not only captured the imagination and the
 excitement of the entire audience and, obviously, we are booked
 out, we are sold out, every seat is filled. So, a fabulous
 presentation, fabulous demos, showing the excitement of this
 industry, showing the outlook into the future and how Intel is
 leading the way in microprocessor process technology as well as
 product technology and I think everybody will be going away from
 this presentation with a much better understanding of the power of
 this entire industry.

 With that, thank you again and let me hand you a memento to this
 important meeting, Innovate 97. It�s not as big as your picture,
 but it�s very nice.

Legal Stuff � 1997 Intel Corporation
    
678.14re: .9 by Yield::HarrisPERFOM::LICEA_KANEwhen it's comin' from the leftWed Apr 30 1997 13:2626
|>   Yes we should never underestimate Intel and what they are capable of
|>   delivering.
|    
|   Is any vendor currently delivering systems with anything better than 
|   200Mhz Pentium Pro chips in them?
    
    Except for a few folks playing overclocking games, no.
    
    Yes the rumors about a 266MHz Pentium Pro last summer were greatly
    exagerated.
    
    BUT!
    
    Intel claims that they still are following Moore's Law (which they
    now define as doubling performance every 18-24 months, a subtle
    change showing up in a recent visual computing white paper).
    
    That means that a 400(+?) MHz Pentium II might be expected any
    day now (for the very paranoid) or by the end of the 1997 (for
    the just paranoid).
    
    
    Or, to put it another way.  That image in the mirror is only
    12 months behind at integer performance.  (But way way behind
    at floating point.)
    								-mr. bill
678.15PCBUOA::KRATZWed Apr 30 1997 14:093
    You should read about DIGITAL's Celebris GL-2 (dual 266 P2) in
    an upcoming (possibly next Monday's) PC Week.
    K
678.16NT SPEC numbers for Digital w/ Pentium IIFORBIN::WILKINSONThu May 01 1997 10:303
    Will we be releasing NT SPEC numbers for the Celebris GL-2/1?
    
    Hugh
678.17PCBUOA::KRATZThu May 01 1997 12:386
    The relabelled SBU box, PWS 266i2, may.  The CSG group is currently
    trying to figure out of they want to publish numbers with Intel
    Reference compiler (good), or Microsoft compiler (not so good).
    SPEC doesn't mean a lot to the PC/NT (desktop Celebris) market...
    having long since moved on to real application benchmarks.
    K
678.18SPEC is composed of *real* benchmarks tooNPSS::WATERSI need an egg-laying woolmilkpig.Thu May 01 1997 13:386
>    having long since moved on to real application benchmarks.

  Many people in DIGITAL and elsewhere are familiar with a majority of the
  SPEC component benchmarks as real applications, used to develop our
  products or run our business.  I think you meant "desktop office
  application", not "real application".
678.19PCBUOA::KRATZThu May 01 1997 14:135
    Well, even in the desktop hi-end NT market, real application benchmarks
    (AutoCAD, CADbench, ProE, Photoshop,...) have pretty much supplanted
    the synthetic loop-unrolled/KAPed "application" benchmarks in terms
    of representing realistic system performance.
    K 
678.20SPEC's have their placeSUBPAC::FARICELLIThu May 01 1997 15:3920
   Oh boy, another p*ssing contest with Kratz!

   OK, I'll bite. The SPEC benchmarks help measure performance across
   a variety of computers/operating systems. They include, if I'm not
   mistaken, SPICE 2G6 (the circuit simulator), a C compiler,
   image compression, etc. We can quibble about whether these are
   "applications" -- they are certainly better than things like
   whetstones and that ilk that proceeded them. Until Alpha,
   HP-PA, MIPS, PowerPC all get crushed by Intel, we'll still need
   processor and OS independent benchmarking.

   SPEC's give you an overview of a system. If you are
   primarily concerned with Wintel apps, then by all means,
   benchmark using those apps. Ditto for databases.

   -- John Faricelli


   
678.21QUARK::LIONELFree advice is worth every centThu May 01 1997 16:416
I've seen some SPEC numbers from John Henning using DIGITAL Visual Fortran
which are quite good.  The SPECbase numbers, in fact, were on a par with or
even better than UNIX.  He hadn't run the "all out" SPECfp tests - NT doesn't
have all the post-optimization tools that UNIX does.

				Steve
678.22Both are useful.TALLIS::GORTONFri May 02 1997 09:2848
    
    Re: .19
    
    Get real.  SPEC has _never_ pretended to have it's benchmark programs
    be representative of entire systems.  EVER.  Unlike the plethora of
    "real application benchmarks" that you mention, SPEC very carefully
    measures specific characteristics of a system.  SPECint9x and SPECfp9x
    measure CPU, memory, and compiler performance.  The "system" benchmarks
    that you talk about all have a very significant graphics component to
    them, which means that poor general CPU performance can be hidden
    behind a snappy graphics card.  Benchmarks are intended to be methods
    for customers to choose systems that are appropriate for them.
    They (benchmarks) simply happen to be used as marketing tools, which
    is why they are hyped a lot.
    
    The history of SPEC is that it comes from a world where people wanted
    to be able to very carefully measure specific attributes of systems
    that they were evaluating for purchase.  The PC marketplace is a
    marketplace for the masses, with customers that often don't really
    understand (for whatever reason) the performance characteristics of
    what they are buying.  And so the benchmarks are of the 'one app to fit
    them all' variety.  A complete synthesis of a lot of components boiled
    down into a single measurement.
    
    As an example, I don't particular depend upon graphics in my day-to-day
    work, so a high-end graphics card would be wasted.  But I can, and do
    frequently use all of the computes and memory in my system.  So when
    my system was upgraded, it went from a Jensen to a 500Mhz PC164 with a
    TGA graphics card, and a full load of memory.  The applications I
    typically use are compilers, and other low-level systems tools. 
    
    How many PC buyers use these "system" and "real application" benchmarks
    but don't really understand how much their particular workloads are
    influenced by the CPU speed, the graphics card, the memory subsystem,
    or the disks?  I would guess upwards of 30%.
    
    The _really_ good thing about seeing SPEC95 performance numbers for
    both a UNIX and an NT system means that for the very first time,
    there is a way to have _some_ comparison (beyond clock speed) of the
    compilers/OSs, and to see that the compilers for Alpha/NT don't suck.
    Frankly, I think these numbers will go a long way to helping the
    credibility of NT as a viable development platform for compute-hungry
    users.
    
    There is a place for SPEC, and there is a place for the "real
    application benchmarks".  SPEC is interesting for customers that
    need to understand the computational characteristics of systems,
    while BAPco, et. al. are useful for less sophisticated end-users.
678.23PCBUOA::KRATZFri May 02 1997 12:008
    If you want volume, you'll push application performance first
    and foremost in the hi-end desktop NT market (and Alpha does
    absolutely magnificent where ported; ProE, etc).
    
    If you want to continue with what you're doing, SPEC has, and
    will continue to, serve you fine.
    regards,
    Kratz
678.24Intel doesn't bash SPEC, why does Kratz?PERFOM::LICEA_KANEwhen it's comin' from the leftFri May 02 1997 13:5430
    Oddly enough I suspect that Intel will publish another performance
    brief next week.
    
    Even more oddly, it will certainly contain SPEC CINT95 AND CFP95
    results.
    
    Actually, not oddly at all.
    
    Intel categorizes CINT95 and CFP95 as COMPONENT APPLICATION BENCHMARKS.
    
    
    They are indeed application benchmarks that measure a subsystem
    of a computer.
    
    
    And last I checked, Intel also publishes seven specific system-level
    application benchmarks such as ProE.
    
    For those seven applications, the application benchmarks are indeed
    critical.
    
    Certainly a ProE user might strongly consider the results of the
    ProE benchmark.  But winning the ProE benchmark probably isn't in
    the mind of a Bryce 2 user when they are selecting a system to purchase.
    
    (Winning ZD benchmarks, either application or synthetic, probably
    does matter to more than a few Bryce 2 users.  Same thing with
    Byte benchmarks.)
    
    								-mr. bill
678.25I'm pro-growth, not anti-SPECPCBUOA::KRATZFri May 02 1997 14:116
    Alpha sales are absolutely pathetic (0.1 market share) and stagnant
    for the last year, Bill.  You seem to have no bashfulness in blaming
    sales and marketing... whaddaya say they try pushing something
    different and see what happens?
    Kratz
    
678.26Pentium II - May 7, 1997PERFOM::LICEA_KANEwhen it's comin' from the leftWed May 07 1997 12:4911
    The Pentium II 233, 266 and 300MHz are announced today.
    
    			233	266	300
    SPECint95		9.49	10.8	11.6
    SPECint_base95	9.49	10.8	11.6
    SPECfp95		6.43	6.89	7.20
    SPECfp_base95	5.84	6.36	6.78
    price (Q1000)	$636	$775	$1,981
    Available		now	now	Q3CY1997
    
    								-mr. bill
678.27RLTIME::COOKWed May 07 1997 15:394

Any specification about the system this was produced in?  

678.28configPCBUOA::KRATZWed May 07 1997 16:5014
    The Pentium II Performance Brief is up on Intel's web site.
    The SPEC95/Unix machine was kind of bland:
    	440FX
    	Pentium II with 512k
    	64Mb EDO
    	Matrox Millennium
    	IDE Quantum Fireball 1080 (curiously not that fast)
    	Unixware V2.0 and the Intel reference compilers. 
    
    Dell is selling pretty much the same hardware for $3899:
    266, 512k, 64Mb EDO, 6.4Gb, 17", 8Mb Matrox, 16x CD,
    wave table sound + Altec speakers, Iomega zip drive,
    33.6k modem. Office97 small business edition, Windows95
    Kratz 
678.29HELIX::SONTAKKEFri May 09 1997 10:375
    What's the timeframe on the availability of 21164PC motherboards?  I
    would like to see apple to apple (with the availability date, prices
    and performance data) comparison.
    
    - Vikas
678.30From Electronics Weekly - 7-May-1997WOTVAX::16.194.64.143::16.195.0.183::bellMartin Bell @BBPFri May 09 1997 11:5328
There is a nice article in this weeks Electronics Weekly, detailing
the "Pentium II debut".

The first half of the article compared the processors with those from
AMD and Cyrix. It says that the prices are $636 and $775 for the 233MHz
and 266MHz respectively (in 1,000 unit quantities).

However, it is the SECOND half of the article that is really interesting!

"The Pentium II fares less well when compared to Digital's recently
announced stripped down Alpha, the 21164PC. The Alpha for desktop PCs is
priced at $495 for a 533MHz part. Digital claims a SPECint95 of 14.3 and
a SPECfp95 of 17. In contrast the Pentium II manages only 10.8 and 6.9
respectively.

Intel plans to release a 300MHz workstation version of the Pentium II
in the second half of the year, priced at $1,981. This is believed to
be codenamed Deschutes, a 0.25um five layer metal shrink of Pentium II.

The current leading workstation processor, Digital's Alpha 21164, costs
under $2000 for a 400MHz part. Intel's 300MHz Pentium II has a SPECint95
of 11.6, compared to 11.7 for the Alpha. Floating point, however, is a
different story. Intel estimates 7.2 for SPECfp95, while the Alpha more
than doubles that at 15.9."

If only we got more press like this!

mb
678.31present vs future comparison...SMURF::STRANGESteve Strange, UNIX FilesystemsFri May 09 1997 14:5311
    > The current leading workstation processor, Digital's Alpha 21164, costs
    > under $2000 for a 400MHz part. Intel's 300MHz Pentium II has a
    > SPECint95 of 11.6, compared to 11.7 for the Alpha.
    
    It's a bit annoying that they're making a comparison between the 21164
    that has been out for a while, and the 300Mhz Pentium II which won't be
    out for a few months.  It would make more sense to compare with 600Mhz
    21164, at the very least.  But better press than we usually get (i.e.,
    better than none)!
    
    	Steve
678.32PCBUOA::KRATZFri May 09 1997 16:034
    And the 400Mhz part was announced at $1913 ("under $2000") last
    *March*... before the December price reductions.  Hopefully
    it's well under $1k now (or, if not, perhaps that helps explain
    things...). K
678.33Apples and OrangesAWECIM::SEGALFri May 09 1997 17:543
They weren't event close. I don't know how much a 400 mhz 21164 goes for, but,
I saw a note that the current price for a 500Mhz EV56 21164 is $995, so one 
would hope/assume that the 400 (or 433) mhz chips are significantly cheaper.
678.34But Which `DECchip 21164'?XDELTA::HOFFMANSteve, OpenVMS EngineeringMon May 12 1997 11:049
   At the risk of harping on this, *which* DECchip 21164? 

	The `DECchip 21164', or the `DECchip 21164'? 

   The more recent -- often incorrectly called the 21164A -- has a
   different internal code-name, EV56, and runs rather faster than
   400 MHz, and has different instruction support, etc.

678.35I'd assume it's EV56WRKSYS::INGRAHAMAndyMon May 12 1997 12:325
I don't think the first `DECchip 21164' (aka EV5) runs as fast as 400 MHz.
Its upper limit had been 300 MHz, which I think was later upped to 333 MHz.

Even where EV5 and EV56 overlap (300-333 MHz), the EV56 is preferred for
new designs because of lower power (among others).
678.36Missing A is really quite dumbDECC::OUELLETTEmudseason into blackfly seasonMon May 12 1997 12:374
TurboLaser cranked the EV5 to 350 MHz (with a bit more power).
At least one of the 3rd party Alpha hardware houses clocked
EV5 at 400 MHz.  And then there was the company with the
refrigerator like cold plate...
678.37.35, .28, or .25?TBLADE::GEEHANOpenVMS and NT; Perfect Together!Tue May 13 1997 17:0018
    
    The Pentium II fact sheet in Intel's home page says that all three
    speeds; 233, 266, & 300MHz are done with .35 micron. The info also
    doesn't mention that the 300 won't be available until Q3CY97 (see reply
    .30).
    
    The speculation was that 'Klamath' 233 & 266 would be .28; and that
    'Deschutes' would be a .25 shrink, 300MHz, possibly 333MHz.
    
    Did Intel get 300MHz with .35, or is the speculation in .30 correct?
    Would that make the 450MHz part the 'Deschutes' at .25? Or is the Intel
    fact sheet wrong? Or maybe just creative marketing?
    
    
    
    							Thanks,
    							djg...