Quo Vadis Intel? — The Latest Roadmap — THG.RU

Introduction

Currently Intel is offering three different processor types and a variety of chipsets for the mainstream market. Interestingly enough, none of the three processor-types is perfect, particularly if you look at them in terms of price/performance. The low-end or ‘value’ processor ‘Celeron’ is unable to deliver the same price/performance as AMD’s Duron processor, the same is valid for Intel’s mid-range processor Pentium III, which has big difficulties to compete against AMD’s Athlon in terms of performance as well as price. Pentium 4, Intel’s new flagship processor, is hindered by its rather inconsistent and in large parts unimpressive performance in today’s applications as well as its high price. To make matters worse for this CPU, it can only operate with the expensive and still rather unpopular RDRAM memory, which even needs to be purchased in two pieces rather than only one.

Still there’s of course no reason to feel sorry for Intel, since this company is still able to sell even inferior products at high prices due to its marketing machine, long history and a large following of OEMs, who always smell big bucks as long as they sail in Intel’s wake.

Intel Influencing The Future

Intel’s roadmaps, which are of course supposed to be kept secretly from the press and the public, are often more than just listings of Intel’s future products. In every roadmap that I have seen so far (and that’s some 50 or more) Intel is in fact foretelling the future. In principle, there’s certainly nothing wrong with that. We all happen to claim we know what will happen in a few days, months or years from now once in a while. However, there is a difference in general perception, depending on who makes those projections and who actually listens to them. Intel has a long and respected history of being able to ‘bend’ the future of the PC-market. Therefore, if Intel tells the OEMs that something is supposed to happen, the OEMs believe it and act accordingly, so that in the end it is indeed taking place. Wouldn’t you sometimes wish you had the same influence on the future as well? I bet AMD does.

Basically, Intel’s roadmaps are supposed to tell the customers what they are expected to do, to make (Intel as well as themselves) as much money as possible, and except for the mistakes made with the 820 chipset, the bad image of Rambus and its expensive but below-par performing RDRAM memory, the following MTH-debacle and the retraction of Pentium III 1.13 GHz, Intel has served its customers well with it. Today, Intel wishes to forget those former failings and tries to pick up the threat of the times before Camino and Pentium III 1.13 GHz. The roadmaps are as full of enthusiasm as ever.

The Current Pentium 4 Situation And Its History

Right now, Intel’s latest and greatest processor product goes by the name ‘Pentium 4’. It was released with huge marketing efforts in November 2000 and is now trying to win over the mainstream and performance market. While the performance of Intel’s new flagship is looking pretty inconsistent, Pentium 4 benefits from the old MHz or today GHz-tale. The average Joe who wants a fast computer believes that a lot of GHz means a lot of performance and since Tom’s Hardware is today only reaching about 2.5 million readers world-wide, this situation won’t change much in the years to come. Combine the GHz-hype with Intel’s history, brand name and marketing machinery and you can sell Pentium 4 even at high prices or bad configurations, as partly pointed out in the recent OEM-system article.

Now this situation is not all new at all. In fact, it has hardly ever been different before. The difference however is that Intel is using Pentium 4 and its hype also to push Rambus RDRAM memory. Intel’s first attempt to do so with Pentium III can be considered as close to a failure. While Pentium III was unable to benefit from RDRAM’s virtues (high memory bandwidth), it was heavily penalized by RDRAM’s failures (high latency). Still Intel managed to introduce RDRAM to the market, ensuring that it would be available once Pentium 4 arrives on the scene. The new flagship processor is now in fact able to benefit from RDRAM’s high bandwidth and this is considered to be good enough, since Intel was wise to ensure that Pentium 4 would not run with any other memory type. While alternative memory types are in fact available, Intel simply decided against supporting them with any Pentium 4 chipset. This means for us reviewers that we cannot compare the performance of Pentium 4 with RDRAM against any other solution and thus we simply don’t know if Pentium 4 might actually benefit from memory with lower latency than RDRAM. Maybe RDRAM is indeed great for Pentium 4, but how is anyone supposed to know unless he’s tested alternative (low-latency) memory types with this processor?

Intel’s RDRAM Credit Program

After Intel’s forced introduction of RDRAM for Pentium III systems, its failure to deliver actual performance and the publicly highly disapproved lawsuit attitude of Rambus Inc., the majority of computer users built up strong reservations against RDRAM. Additionally, RDRAM still happens to be rather expensive when compared to other memory solutions. Thus Intel had to do a bit more than just forcing RDRAM upon people interested in Pentium 4.

The other step to ensure success of Pentium 4 and thus RDRAM is called ‘RDRAM Credit Program’. Since the early days of Pentium 4 OEMs and system integrators received a $60 rebate per sold Pentium 4 processor, to cover the price premium of RDRAM memory. This program continues until April 15, 2001, when it is switched down to only $30 per Pentium 4 processor, and only for those that are running at 1.4 GHz or more. The final end is supposed to be on May 27, 2001, the time when Intel expects RDRAM to be inexpensive and accepted enough to become just another memory type.

You wonder why this program never existed for Pentium III processors. My own guess is that Intel knew very well indeed, that Pentium III would not perform well with RDRAM. The whole Rambus galore of the last 15 months was only there to introduce this memory type to the PC-market. While we technological publicists shouted «RDRAM is a fraud! Dont’ buy it!» the managers at Intel were silently nodding, smiling and thinking «We know, we know, but you will finally accept it, as we know what comes next and you don’t«.

The Brookdale Chipset

Some of the more experienced readers will now say ‘What is his problem? Doesn’t he know that Intel will finally launch a Pentium 4 chipset with PC133 and DDR-SDRAM support?» Yes yes, I am actually aware of that. However, Intel is deliberately delaying this alternative platform until it has ensured that RDRAM will be the widely accepted memory type for the ‘net-bursting’ Pentium 4.

According to the latest roadmap Intel will release the code named ‘Brookdale’-chipset in Q3/2001, but at this time it will only support the good old PC133 memory, which offers only a third of the memory bandwidth found in the current i850-chipset with its dual-RDRAM channel configuration. Hardly anyone with performance in mind will be interested in this PC133-Brookdale chipset by then. Finally Brookdale will actually get DDR-memory support, but not before the first quarter of 2002, which is more than one year after the release of the first DDR-memory supporting chipset for AMD’s Athlon processor. It seems preposterous to think that Intel isn’t able to finish the design of Brookdale a little bit earlier. Remember the more or less secret agreement between Rambus and Intel. In two following quarters Intel has to sell more RDRAM-chipsets than chipsets with support of alternative memory types and it gets Rambus shares worth hundreds of million dollars. Is there any easier way to achieve that than by delaying alternative platforms for Pentium 4? Intel does not necessarily offer what is best in terms of performance, it sells what is best for its business, even if it doesn’t seem to make sense. Remember the, on the first look pointless, release of i820 and i840. We know now that Pentium III is simply unable to benefit from RDRAM. I can’t imagine that Intel’s engineers were unaware of this. Why would Intel release a chipset that hinders the performance of its processors?

‘Brookdale’s’ preliminary specs:

Northwood

It is pretty well known to most of you, that the current Pentium 4 processor will soon be replaced by a more advanced version. You will also be aware of the fact that this new version won’t be compatible with current Pentium 4 platforms, making current Pentium 4 systems anything but future proof.

The upcoming Pentium 4 is found in Intel’s roadmaps under the code name ‘Northwood’ and is supposed to start replacing the ‘old’ Pentium 4 in Q3/2001. Besides other performance enhancements it will come with 512 kB second-level cache and therefore twice the L2-cache of today’s Pentium 4. Northwood will be produced in 0.13 micron process and it will have a new package for a new socket called mPGA478, for it will have 478 pins, quite a few more than the 423 pins found in the Pentium 4 of today.

Please always keep ‘Northwood’ in mind when considering the purchase of a Pentium 4 system today. This system will be old news without an upgrade path once ‘Northwood’ starts replacing ‘Willamette’ in the second half of this year.

The Future Of Pentium III — Coppermine / Coppermine-T / Tualatin

Enough talk about Pentium 4! Pentium III is still found in the most systems out there and Intel sees Pentium III still as the ‘volume leader’ in 2001. Right now, Pentium III is suffering from two (partly only theoretical) problems. In terms of clock speed it can’t reach AMD’s Athlon, which will soon be available at 1333 MHz, and in terms of price it lags behind its competitors from AMD as well. Intel is planning to move the pricing of the currently known Pentium III models into the ‘Mainstream 1’ segment, which is one step above the ‘value’ segment. In other words, Pentium III will become less expensive and this seems like a wise decision, because it is hard to find a reason for the purchase of this processor right now.

The next-generation Pentium III goes by two different code names. The real new version of Pentium III is called ‘Tualatin’. It will be produced in 0.13 micron process and reach clock speeds of 1.26 GHz and possibly beyond. ‘Tualatin’ will utilize lower voltages and be incompatible with current Pentium III platforms. Besides ‘Tualatin’ there is ‘Coppermine-T’. This processor-model is still made in 0.18 micron process, but it is compatible with Tualatin. It doesn’t seem clear if ‘Coppermine-T’ is simply a ‘Coppermine’ that can run at the new voltages of ‘Tualatin’ or if it will indeed be a new processor design.

As ‘Coppermine-T’ seems rather nebulous right now, I prefer to concentrate on Tualatin. The interesting thing with this new Pentium III core is the fact that it will exist in two versions, one with a 256 kB L2-cache and one with 512 kB L2-cache and thus twice the second level cache size of current Pentium III processors. In the current roadmap you can find Tualatin with 512 kB L2-cache only in two segments. The mobile Tualatin for notebooks will actually come with the larger second level cache and then there seems to be a desktop Tualatin for dual-processor systems that will also have the 512 kB L2-cache. All the normal desktop Tualatin Pentium III processors are supposed to only have the ususal 256 kB L2-cache.

There isn’t much known about Tualatin’s specs, besides the second level cache size, the lower voltage and the different manufacturing process, but rumor has it that Tualatin will actually use a faster front side bus, possibly the quad-pumped 100/400 MHz bus of Pentium 4, enabling it to benefit more from high bandwidth memory. This rumor seems rather sketchy however.

Almador Only for Notebooks?

Fact is that Tualatin will require a different chipset north bridge to accommodate the changed interface. The so called ‘B-step’-version of i815 will be able to run with Tualatin and probably Coppermine-T. In previous roadmaps used to be also the ‘Almador’ chipset for Tualatin, but the latest roadmap doesn’t list it in the desktopn section at all. It does turn up in the mobile roadmap however, as ‘Almador-M’ or 830M chipset. It seems as if Intel dropped the Almador chipset in its desktop plans.

It is not yet known if i830 will indeed support DDR-SDRAM, but it seems rather likely. Besides this feature and the new ICH3 south bridge there isn’t much difference to the already known i815 chipset. Almador / i830 will also come with integrated 3D-decellerator.

Intel Roadmap Step by Step

Let’s now go through the roadmap systematically.

Intel Roadmap — General Statements

• Pentium III will be volume leader in 2001.
• Pentium 4 is supposed to reach 50% of the performance/mainstream platforms by Q4/01.
• The transition Socket423 to Socket478 (Northwood) will begin in Q3/01.
• Brookdale with PC133 SDRAM support will come in Q3/01 = high confidence/low cost (P4 supports ‘all mainstream segments’ by that time)
• Tualatin will start to enter value segment in Q2/02.

Intel Roadmap — Desktop

Intel Roadmap — Mobile

Intel Roadmap — Workstation / Dual Processor Server / Multi Processor Server

Explanations

All those code names are a bit confusing and I didn’t feel like writing about each product. However, I want to give you a few comments that are hopefully helpful to understand the whole enchilada.

• ‘Northwood’ will be the next-generation Pentium 4 processor, which will not be backwards compatible with previous Pentium 4 platforms. Northwood is expected in Q3, 2001.
• ‘Tualatin’ is the next generation Pentium III processor. It will also require a new platform, either the B-step i815 chipset or i830 Almador. The latter seems to have been dropped for desktop systems however and might only appear as i830M in notebooks.
  
  The normal Tualatin for desktop systems is supposed to come with 256 kB second level cache, while the notebook version will have 512 kB. However, there is supposed to be a Socket370-version of ‘Tualatin-512’ available as well, targeted to dual processor systems.
  
  The roadmap says that Tualatin will be priced higher than Pentium III, but also higher than some Pentium 4 processors. Prices for Tualatin are not available yet, but are expected for March.

• ‘Foster’ is the workstation/server version of the Pentium 4 that we know now. It will be supported by the i860 chipset, which is a workstation version of Intel’s 850 chipset. Later there will be Foster processors with integrated third level cache for multi-processor systems. I expect that Foster will receive the official name ‘Pentium 4 Xeon’.
• ‘Prestonia’ is the ‘Xeon’-version of ‘Northwood’ and thus the next-generation workstation/server processor after ‘Foster’. It will also be manufactured in 0.13-micron process and it might also require a different socket than ‘Foster’.
• The ‘Plumas’ chipset will probably be the workstation/server version of ‘Brookdale’, including DDR-SDRAM support, but also dual processor and maybe even MP-support.

So that’s it, Intel’s latest roadmap. I hope you enjoyed it.