The first half of this year has every right to be called Intel's worst period. The launching of allthe new products was postponed for some reasons throughout the whole year. Among those were the so longawaited processors on the new Coppermine (CuMine) core and the revolutionary i820 chipset also know asCamino supporting the new memory type - Direct Rambus DRAM. There is nothing eternal under the moon andthe dark time had to be replaced by a light stripe. Intel engineers have been summoning up fresh energyfor a significant while already and autumn is definitely high time they showed their worth.

In the present article we will try to lift the curtain and to take a look at Intel's possible futureplans in order to answer numerous acute questions. So, having undergone the whole bunch of differentchanges the current roadmap version looks as follows (without server and entry workstations market,focusing mainly on the items, which are of interest to average users, namely on Mainstream desktop andcheaper Value PC):

Here it is on purpose that we omitted concrete dates, because we wouldn't dare to state somethingfor sure knowing about Intel's gift towards very discouraging actions. Nevertheless, we still have acouple of exact dates right at hand, but let's start from the very beginning.

27 September 1999 - 133MHz FSB's Day

Well, this will be the starting day for the systems with 133MHz system bus. We all know that VIAlaunched its Apollo Pro133 chipset, which allows using this system bus working frequency, quite a longtime ago already, however there haven't been made any processors supporting this FSB that's why fullyfledged 133MHz are still a fantasy. In the end of September the situation should undoubtedly change -the world will see the first processor designed for this FSB frequency. Unfortunately, it is not goingto be Coppermine, as all of you have probably expected. Just to make sure you remember: Coppermine isa Pentium III made with 0.18 micron technology and provided with a 256KB L2 cache integrated into theprocessor core and working at the processor frequency. The developers made too many mistakes whiledesigning this core that's why there is no chance for us to see the processor on it in the end ofSeptember, as we have expected before. This cool event is postponed till the last quarter of this year.

Therefore those of you, who look forward to enjoying the slight performance gain they will get dueto the increased FSB frequency, will have to agree to a Pentium III on the old Katmai core supportingnew FSB. On 27 September they will announce processors like that - Slot 1 Intel Pentium III 533 and600MHz with 133MHz FSB and an external (relative to the core) 512KB L2 cache working at half theprocessor frequency.

However, Intel won't be satisfied having launched only a few new processors. In this case thesituation could appear quite awkward: all the novelties could be supported only by mainboards based onVIA chipsets. The same day is going to be also marked with the coming of two new chipsets - i820 andi810e. The launching of i820 is a kind of epochal event: this chipset was postponed and redesigned acountless number of times (at first the official launching day was planned for this June). But finallythe developers and the potential customers have reached mutual agreement on the matter and as a resulti820 may now enter the chipset market.

Intel 820

The problems with this chipset appeared, because of the great number of different new technologiesIntel tried to introduce at the same time with one single product - i820 chipset: Direct Rambus DRAM,AGP 4X, 133MHz system bus frequency. As for VIA, which is also undertaking some steps in this direction,it seems to be more careful. Now they launched a chip supporting 133MHz system bus, a bit later they aregoing to offer its AGP 4X version and then in a few months they will also introduce DDR SDRAM support.There is no need to say that gradual, step-by-step, launching of new products makes it much easier forthe manufacturer to eliminate the occurring defects or errors.

The greatest advantage and at the same time the greatest drawback of i820 is an absolutely new forPC memory type - Direct Rambus DRAM. In general, it is the most doubtful innovation - the first i820based mainboards will force the users to completely shift from the memory type they used before(DIMM modules) to the new ones (RIMM modules). And if you take into account their high cost and thehuge memory size required by the today's applications, the whole purchase will make a crazy sum, whichis very unlikely to arouse public delight.

But though the situation looks just hopeless, Intel resorts to its main trump - highly effectiveAGP 4X bus supported by the new chipset and hence requiring fast system memory. It's really true: thebus providing peak bandwidth of almost 1GB/sec, 133MHz processor with the same bandwidth, together withsuch trifling things as PCI bus with the bandwidth of 132MB/sec, make its valuable contribution to thepeak bandwidth required from the memory, which has to be about 2.1GB/sec. No doubt that PC100 SDRAMwith its 800MB/sec has nothing to do here.

However, the graphics cards and PC133 SDRAM manufacturers offer another argument in their turn. Theystress that the peak bandwidth and that active most working time are a bit different things. And asystem like that will have a real bandwidth of 500-600MB/sec, which could be easily provided even byPC100, not to mention PC133 SDRAM. This fact got proven by the preliminary tests of i820 basedmainboards - the applications were run at the same speed they worked on i440BX. Of course, thesituation may change as soon as a new generation of graphics accelerators comes out, but for now thethings stand as they are: Intel tries to run before it learns to walk.

Well, what will this i820 actually be? First of all, this chipset will be based on a new AcceleratedHub Architecture (as in i810) supporting 100 and 133MHz processor bus. Besides, we have alreadymentioned AGP 4X and Direct Rambus DRAM support. And the last thing we should also point out is thelist of features we saw by i810: Ultra ATA/66, AC97 compatible codec (AMR slot) and chipset integratedhardware random number generator.

It looks as if high prices on DR DRAM would make Intel launch two i820 versions. The first one willbe intended for highly productive systems and will support PC800 and PC700 system memory, which worksat 400 and 356MHz correspondingly (Rambus implies that the data is transferred on both signal fronts).And the second version will be somewhat simpler and will support only PC600 Rambus memory working at300MHz. But it is not the end. Trying to satisfy all the tastes Intel made its i820 support the goodold PC100 SDRAM memory as well. However, Intel's ideology won't allow them to make their offspringsupport good new PC133 memory. For this purpose the mainboard manufacturers will be offered anadditional hub-microchip - MTH (Memory Translation Hub) with PC100 SDRAM interface. In the end, besidestwo RIMM slots, which may allow installing Direct Rambus DRAM, i820 based mainboard may also acquire2 DIMM slots.

On the other hand, theoretically it is quite possible that some mainboard manufacturers will beable to make use of the MTH by a third company, which will also provide 133MHz SDRAM interface, -the situation similar to the way they started using other clock frequency generators having overclockedBX chipset far beyond the nominal 100MHz. However, it happened not quite right away after its launching.

The second chipset - i810e - also aims at using 133MHz processor bus, but it is intended for the mostlow-cost systems and hence doesn't support DR DRAM. In other words, this integrated solution is noneother than an improved i810 with almost 15% higher graphics subsystem performance. This performanceincrease became possible mostly due to 133MHz display cache support in i810e. Here all the significantdifferences between i810e and i810 are over. Unfortunately, there is absolutely no hope that Intel willever support PC133 SDRAM in its new products we are talking about.

Nevertheless, on 27 September Intel will move from the dead point and pretty soon it will please uswith another set of novelties. Let's wait till October.

25 October 1999 - Coppermine's Day

0.18 micron technology into life! This slogan is the best way to describe the events, which areplanned for 25 October. This day will be the beginning of the mass sales of Pentium III processorsmade with a new technology and stuffed with a new core - Coppermine. 256KB L2 cache working at the corefrequency and similar to that we have in today's Celeron promises a new performance upsurge. Besides,Intel has also optimized the work with L2 cache, which will also contribute to this performance rise.

Coppermine doesn't require any external L2 cache microchips, which forced Intel engineers to providetheir Pentium II and Pentium III with a special cartridge. The whole thing was assembled on a plant inIreland reequipped especially for this purpose. With regard to this fact the novelty may see the lightin both existing form-factors supported by the today's mainboards - Slot 1 and Socket 370.

Slot 1. By this day we will already be using the systems with 100 and 133MHz system bus, so the newCoppermine cores, as well as Katmai ones, will also be available in two versions: for 100 and for 133MHz.On 25 October we will get five different options: 600/100, 650/100, 600/133, 650/133 and 667/133MHz(note: not 666 but 667 - Intel seems to believe in superstitions ;-).

As for Socket 370, the situation is more interesting. Some time ago it was still planned that PPGACoppermine, or FC-PGA (Flip-Chip) as they are called in this case, will use a new but backward compatibleSocket 418, which will differ from Socket 370 by an additional internal row of pins connected to theground or power. This was made because of the stability problems of 0.18 micron crystals packed into PGA.

FC-PGA Coppermine Processor

However, a bit later Intel managed to overcome this trouble and decided to give up Socket 418.However, for the time being it had to sacrifice SMP-configurations support and 133MHz system busfrequency. As a result, the first Socket 370 Coppermine will be only 100MHz and uniprocessor. On 25October 500 and 550MHz Socket 370 Coppermine will come out. Although they will be not as cool in termsof frequency as their Slot 1 brothers, they will still be quite normal Pentium III ones, and notCeleron.

But you should bear in mind that since all these new processors are made with a thinner technology,their core voltage will be less than that used today: supposedly 1.6V instead of the usual 2V. And itis exactly here that the main problem crops up: compatibility with the older mainboards. We would liketo devote it a bit more time because it has to do not only with the support of the mainboard VRM withthe corresponding voltage, but also with the GTL+ system bus. For the first time in the history of x86systems this bus will have higher voltage than the CPU, which will require some special circuits on themainboard. Well, below you may see the compatibility table of the mainboards and new chipsets:

We have also included in our table the mysterious Coppermine 128K we are going to speak about later.Besides, we also feel like commenting on a couple of things here. It's quite evident that BX/ZXmainboards don't support any 133MHz processors because they can't provide the AGP bus with the needed66MHz at this frequency. The second thing, which may cause incompatibility, is the different voltage ofthe GTL+ system bus and the processor core. And in case of Socket 370, there appears one more"but", namely absolutely different functions of the five FC-PGA contacts compared to those ofthe old PPGA processors. That's why if you need to check if the old mainboard and the new processor canwork together without any conflicts, Intel thinks you'd better turn to the mainboard manufacturer.However, we would say the following here. Some relatively new mainboards (designed this year) based oni440BX/ZX supporting under 1.8V core voltage, are very likely to support Coppermine, and the older onesaren't. By the way, you should always keep in mind that the processor working at the voltage exceedingthe nominal one will considerably faster run out of its resource that's why all the new mainboards willcome out with a new VRM 8.4 allowing the voltages from 1.3 to 2.05V. The second remark deals with i810.Please, remember that its A2 revision supports only 66MHz Celeron processors because of the numerouserrors, which were found in the earlier chipset versions.

And then?

Well, it's rather hard to say what will happen later because our favorite company is so famous forits flexible behavior ;-), however, we may try to describe a few tendencies, which seem more likely totake place.

As for the processors, the performance increase is not the only thing we should look forward to.Another Coppermine baby - a new Celeron based on this core - is going to be born somewhere about thefirst quarter of the year 2000, if 1 January is not the end of the world :-). This CPU will be made onlyin FC-PGA form-factor and will contain the same 128KB integrated L2 cache. The main differences fromtoday's Celeron processors will be the 100MHz system bus and the long awaited Intel's SSE SIMDinstructions support.

After that there will be a short break till Q4 2000 during which we will hardly witness any importantevents for Intel x86 processors. And in the fourth quarter Intel will introduce its new processor core -Willamette, which will provide stable and reliable performance at the frequencies over 1GHz the existingarchitectures will hardly be able to surpass. Willamette will have 256KB L1 cache and at least 1MB L2cache. Besides, this CPU will be manufactured with 0.18 micron technology and a bit later Intel willshift to 0.13 micron and copper technology, which Intel considers not worth being introduced in 0.18micron unlike its main competitor AMD.

Willamette will use the new Socket 423, and before that Intel will give its heart to Socket 370,though it seems a bit strange, don't you agree? Only in the very beginning the amount of availableSocket 370 Coppermine processors will be limited, later on they are expected to completely replaceSlot 1 versions in the market. That's why if you are planning to upgrade your system, don't forgetthat Socket 370 is a more promising solution, which will live far longer than Slot 1.

And now let's try to imagine what's going on with the chipsets. One day it suddenly occurred toIntel that it can't boast a chipset solution for Value PCs with an external graphics card. In orderto eliminate this omission (which VIA was almost ready to do with its low-cost solutions) Intel engineersdesigned a new chipset under the codename Solano, which is due somewhere in the second quarter of theyear 2000. For now this product exists only on the paper, but nevertheless it looks quite attractive.It is supposed to support AGP 4X, 3 DIMM slots and 512MB memory, AC97 and ATA66. Probably Intel willdare use PC133 SDRAM in this chipset as well, however, in the meanwhile it is sealed mystery. Despite theexternal AGP port, there will also be an integrated AGP 2x graphics core, which can be expanded withAIMM (AGP Inline Memory Module) modules and activated when no card is installed into the AGP slot.

The second interesting thing, which is about to happen in the second quarter, is Timna. It (and notthe chipset :-) slightly reminds us of Cyrix MediaGX, because it is a Pentium III processor with anintegrated 128KB L2 cache, graphics controller and memory controller supporting Direct Rambus DRAM.Timna is expected to be manufactured also in FC-PGA, which will fit into a new socket - PGA370-S.However, it is just a project without any silicon realization, so everything may still change severely.

Mainsteram chipsets will also keep pace with the time. In Q2-Q3 of the year 2000 Camino-2 will be due.This chipset, an enhanced i820, will be specially optimized for Coppermine. Besides all the usual thingsit will also have a four-port USB controller, Ultra ATA/100 controller (we wish we knew what it is :-),multichannel AC97 codec, integrated LAN-controller and a few other features, which will be a commonthing in VIA chipsets by that time.

In other words, there will hardly remain any chance to get bored. And don't forget about AMD with itsown Napoleon's plans :-)!