Introduction
I collected several tidbits of AMD’s and Intel’s plans until the end of this year. Some of the stuff you might have found elsewhere already. I only intended to compile all this information together to give a good overview about what is happening in the x86-processor market within the next 8 months, as far as it is fathomable right now.
Intel’s Woes
Before I will get into the details of Intel’s latest plans, I would like to pinpoint another important issue. On Monday, April 17, 2000, Intel will most likely announce that they will not be able to supply the market with enough processors at least until June/July 2000. Talking to vendors in the PC market reveals a formerly unusual picture. Regardless if you are speaking to motherboard makers, system integrators or OEMs, they all complain about Intel’s inability to supply processors. This may sound pathetic to the normal user, but it is a rather serious situation to the above-mentioned vendors. Each of those companies has made forecasts of how much product they are going to produce and ship and of course they are trying to meet them. Those forecasts seem to get destroyed by Intel’s failure in supplying enough CPUs. A motherboard maker cannot sell motherboards, if the customer cannot get a CPU for it. The same is obviously valid for system integrators and OEMs as well.
Only 60% of the Demanded Processors …
Several sources told me that Intel is currently only able to satisfy about 60% of the demand. The result for the vendors is pretty simple. They will either lose money or they have to move over to platforms for AMD’s Athlon processor. This puts AMD in a great position, because they can easily take advantage of Intel’s problems. At the same time AMD’s Athlon is anyway cheaper and performing at least as good as Intel’s processors. The only problem could be the chipset-supply for Athlon-based systems. Currently there is hardly an alternative to VIA’s Apollo KX133 chipset. The vendors as well as AMD can only hope that VIA will be able to supply enough of those chipsets.
Giga-Trouble
There’s one situation where Intel is looking particularly sad. After AMD had released their Giga-Hertz-Athlon last month, Intel had to follow suit and launched their Giga Hertz Pentium III two days later. Intel had to show the world that it was able to make Giga-processors as well. However, while AMD is shipping Giga-Athlons at rather nice quantities, Intel is not even able to supply evaluation samples of their Giga-CPU to OEMs and system integrators. The few Giga-Pentium III processors out there require a heat sink of gigantic size, featuring a special copper ‘heat pipe’ that leads to another heat sink outside of the computer case. This heat sink costs some $50 – 80, adding nicely to the price of the – anyway unavailable – Giga-PIII. For me this Giga-hoax is exemplary for Intel’s current situation. Over are the days when Intel used to build up large quantities of processors before they were launched. Gone are the times when Intel was the reliable partner that always used to be able to supply product. The OEMs, system integrators and motherboard makers must be close to tears remembering the history when doing business with Intel resulted in certain success and huge revenues. Can you still remember Intel making fun of AMD about their inability to supply K6-processors? Last year before the Athlon-launch Intel used to respond to questions how they see the superior performance of the upcoming new AMD-processor with words like ‘yeah, well, the performance may be ok, but we know that AMD will hardly be able to ship those processors reliably‘. AMD has all reasons to laugh and celebrate. Intel is currently heading at the speed of light for making a huge fool out of itself and AMD is there to save the day. It reminds me of an old line of Buck Murdock (William Shattner) in ‘Airplane II – The Sequel’. “Irony can be pretty ironic sometimes”
Intel’s CPU plans for 2000
- Slot1 is supposed to die a slow, but certain death. Now that Intel’s processors don’t require an external L2-cache anymore, there’s hardly any reason for the cartridge-version (SECC2) of their CPUs, which is fitting in the good old Slot1. In the second quarter of this year Intel wants to ship 50% of their processors in the FC-PGA version for Socket370. Slot1 is not supposed to host any processors that run faster than 1 GHz and there might be a price premium for the Slot1-versions of Pentium III in the second half of this year.
- In the third quarter of 2000, Intel is planning to launch the Pentium III at 933 MHz. Currently there don’t seem to be any plans of supplying 900 or 950 MHz version for the 100 MHz FSB. Supposedly there is a problem to offer multipliers of x9 and x9.5. I consider this as pure hoax though. It might well be that Intel will launch Pentium III 900 and 950. The multiplier is not an issue at all. Celerons can do x9 (Celeron 600) already and x9.5 is not far off as well. The setting for this frequency could e.g. be the old x4, x4.5 or x5 setting.
- ‘Willamette’, the highly discussed successor of ‘Coppermine’, is supposed to be launched at 1.3 and 1.4 GHz in Q4/2000. 1.5 GHz and above is expected for the beginning of 2001. One of the discussions I’ve seen recently is going on about the so-called ‘MHz-gap’ between the ‘Coppermine’ at 1 GHz and the ‘Willamette’ at 1.3 GHz. I don’t see why this should be an issue worth going on about. Who can tell me if ‘Willamette’ at 1.3 GHz is even going to be faster than ‘Coppermine’ at 1 GHz? There are enough people who seriously doubt that, including yours truly.
- The latest Celerons at 566 and 600 MHz are finally using the ‘Coppermine’ core as well, but one half of its L2-cache is disabled. This procedure in combination with Celeron’s pathetic 66 MHz FSB is supposed to make sure that there will still be an artificial performance delta between Celeron and Pentium III. To me this is as if Chrysler would sell a cheap Viper that is missing 5 sparc plugs and using narrow tires. Intel might change their 66 MHz FSB policy once AMD releases its ‘Spitfire’ processor though. What else can they do?
Currently Intel’s plans are not looking like it though. Celeron 633, 667 and 700 are supposed to be launched in Q4/2000 and Q1/2001 is supposed to bring us Celeron 733 and above.
Intel’s Chipset Plans for 2000
- In June 2000 Intel will bless us with its ‘ICH2’ chip. This ‘south bridge’ or ‘I/O Controller Hub 2’ will come with an ATA100 interface, offering up to 100 MB/s data throughput for IDE hard drives. It will also supply 4 USB hubs, and a 6 channel AC97 Codec for the ones that like using the low-cost AMR-solution for sound. Last but not least ‘ICH2’ will sport an integrated LAN, which requires an additional ‘PHY’ chip though. Combining ‘ICH2’ instead of ‘ICH’ with i820, i840 and i815 will make i820e, i840e and i815e out of them.
- The most anticipated Intel chipset this year is clearly ‘i815’ or ‘Solano’, actually called ‘i815e’ or ‘Solano2’ if ‘ICH2’ is used with it. ‘Solano’ will be Intel’s first chipset with PC133 support, thus marking the strongly demanded alternative to Intel’s doomed Rambus-chipsets i820 and i840. Solano will support all the other goodies of Intel’s latest chipset generations, like AGP4x and the ‘Hub Architecture’ and it will come with an integrated 3D-decelerator as well. This integrated graphics solution can be turned off however, and an external AGP-graphics card can be used with it instead. The expectations are high for ‘i815’, since it should definitely perform better than the highly regarded ‘440BX’ chipset and thus outperform any Rambus-chipset. However, Intel would contradict itself if their low-cost i815 should make the glorious i820 and i840 look bad. Let’s see what the outcome will be. Intel can only lose. Either the chipset performs great and Intel proves that RDRAM is obsolete, or the chipset performs bad due to deliberately integrated speed impediments, but then nobody will want it.
‘Solano’ is supposed to be launched in June2000, but Intel’s official words about its availability reads like this: “limited supply for distribution channel.” i815 is not supposed to be in ‘full production’ any time before Q4/2000. This way Intel is trying to make sure that even the last BX-supporter has forgotten about the virtues of PC133 SDRAM. - I wasn’t quite sure where to put ‘Timna’, and it might belong in the ‘CPU’-section, but I hope you’ll read about it here as much as you would have in the above chapter. This combination of a Celeron processor with Coppermine-128 core and the ‘GMCH’ is obviously targeted to the low-cost segment. ‘GMCH’ is probably standing for ‘Graphics and Memory Controller Hub’, and it illustrates ‘Timna’s’ special design integrating a processor, graphics and memory controller in one chip, known as ‘system-on-a-chip solution. Timna will fit into a socket called ‘Socket370S’, obviously using more pins than Socket370. Combining Timna with ‘ICH’ or ‘ICH2’ will make a complete system while keeping the costs extremely low. There’s one funny flaw in this chip though. Originally Timna was supposed to work with RDRAM, but the horrendous price tag for this rather unpopular memory type is way too high for a low-cost solution. Thus Intel announced that Timna will work with the rather cheap PC100 SDRAM, but I am not quite clear if this will require an external ‘MTH’ memory translator hub and thus adding to the system costs, or if the ‘MTH’ will be built into Timna. Whichever it may be, it doesn’t make any kind of sense, as we all know of the dramatic performance loss if the data stream has to be translated into the Rambus protocol and then back into the SDRAM ‘protocol’.
Timna is supposed to ship in September 2000 at speeds of 600 and 667 MHz, obviously using the 66 MHz FSB internally, which comes across particularly laughable. 700 and 733 MHz versions are expected for Q1/2001. - Last but not least there is i850 or ‘Tehama‘. This chipset builds the platform for Intel’s upcoming super-processor ‘Willamette’. It supports Willamette’s new ‘Socket423’, 400 MHz FSB (quad pumped 100 MHz), two Rambus channels and PCI64/66. Its release will obviously go hand in hand with ‘Willamette’s’, so expect it in Q4/2000
AMD
You have certainly heard about it, the last quarter (Q1/2000) was AMD’s most successful quarter in history. Intel’s inability to supply CPUs and the excellent performance as well as availability of the Athlon-processor has finally ended AMD’s long line of losses. An earning of $1.15 a share was more than analysts expected of Intel’s arch enemy and it shows how well AMD is currently doing. This is not all however; the really successful days of AMD are yet to come, because the future processors look even better than what we’ve seen of the well-established Athlon so far.
At the same time there have been a lot of changes at AMD that I personally don’t really appreciate. Many important and good people have left this company in the last months, starting with the departure of Atiq Raza, AMD’s former COO. I will dearly miss men like Dana Krelle, AMD’s Ex VP of Marketing and Lance Smith, former Director of Technical Marketing. Both of those two excellent Ex-NexGen executives seem to have been seriously disgruntled by AMD’s changing internal policy and I wonder if AMD will do well without them. One thing that I have noticed bitterly is the demise of AMD’s dealings with the press. While in the past AMD was a lot more pleasant to deal with than Intel, it now seems as if the new success went to the heads of the few remaining old executives of this chipmaker. Today it’s a lot more fruitful and enjoyable to deal with Intel than it is to deal with AMD. Times have changed, but not all to the better.
AMD’s Processor Plans for 2000
- After permanent changes in schedule, ‘Spitfire‘ will seemingly be launched sometimes in late May or early June this year. This chip is supposed to be AMD’s direct answer to Intel’s Celeron, but it is expected to perform a whole lot better than Intel’s low cost chip. In fact, Spitfire may even surpass the performance of the current Athlon-processors.
Spitfire is an 0.18 micron aluminum-interconnect die with 64 kB of integrated full speed L2-cache adding to a shrunk and optimized Athlon core. Although the L2-cache size is only a eighth of the L2-cache of current Athlons, it may perform even better due to its 256-bit data path and the fact that it runs at full core clock instead of the 1/2, 2/5 or 1/3 speed of Athlon’s external L2-cache. Seemingly, Spitfire will only be manufactured in AMD’s Austin Fab25.
This processor will not be available as a SlotA-version, but is supposed to only ship as SocketA-solution. This socket is AMD’s way to get away from the cartridge-solution that is becoming just as obsolete as Intel’s Slot1, since there is no need for external L2-cache chips anymore. - ‘Thunderbird’, AMD’s successor of the Athlon, as we know it now, will probably be launched at the same time as ‘Spitfire’, to make sure that the low-cost ‘Spitfire’ chip won’t perform better than the ‘high-cost’ Athlon solution. I don’t know what name ‘Thunderbird will get, but ‘Athlon II’ would be a viable solution, don’t you think? Thunderbird is supposed to start at speeds of no less than 1.2 GHz, but it wouldn’t surprise me if AMD will supply slower Thunderbirds as well, replacing the older and slower Athlons.
‘Thunderbird’ is still good for several mysteries. Supposedly there will be a version that is using 0.18 micron aluminum interconnect die with 256 kB of integrated full speed L2-cache, produced in Fab25. At the same time the Dresden Fab30 will supply a 0.18 micron copper-interconnect die. I wonder if this die won’t sport more L2-cache than its brother from Austin, because I have some information saying that there will also be a Thunderbird with 512 kB on-die L2-cache. Whatever it may be, Thunderbird will be right up there competing against Intel’s upcoming ‘Willamette’ and I recon that it won’t have trouble to do so. A shrunk and optimized Athlon core with 256 or 512 kB on-die full speed L2-cache will make today’s Athlons look rather pale, and – who knows – may give Willamette a good run for the money as well.
The first Thunderbirds will come as SlotA version, soon followed by versions for SocketA. - Last but not least there is ‘Mustang’. This chip will be AMD’s entry in the high-end workstation and server market. It’s also based on a shrunk and optimized Athlon core, combined with large 1-2 MB on-die L2-caches. Don’t expect this processor any time before September of this year, because AMD needs to supply an SMP-chipset for it as well.
There’s not much data about Mustang available right now, but I could imagine that this high-end Athlon will be produced in Dresden’s Fab30, taking advantage of the 0.18 micron copper-interconnect technology used in this fab.
It seems as if Mustang will only be available as SocketA-solution, but this information could be faulty.
AMD’s Chipset Plans for 2000
- ‘AMD760’ is the name of the next generation AMD chipset for Athlon processors, and it will support SlotA as well as SocketA. The release date seems to be in September, but I guess that AMD will try to release this chipset earlier if they should be able to do so.
Its features read like this:- Support of 200 and 266 MHz FSB (100/133 MHz double-pumped)
- PC1600 / PC2100 memory = 100 / 133 MHz DDR-SDRAM
- AGP4x
- ATA100 (?)
- SlotA/SocketA
- Dual Athlon support, additional north bridges can be combined with AMD’s ‘LDT’-bus to create systems with 4, 6, 8 CPUs as well. Each north bridge can host 2 Athlons.
- Support of 200 and 266 MHz FSB
- PC1600 / PC2100 memory = 100 / 133 MHz DDR-SDRAM
- AGP4x
- ATA100 (?)
- SlotA/SocketA
You might have noticed that I neither commented on Intel’s Itanium processor, nor on AMD’s upcoming ‘Sledgehammer’. Both CPUs are supposed to mark a new era for the PC. Itanium is a completely new 64-bit architecture that doesn’t have much in common with the 32-bit x86 processors we know today. AMD’s ‘Sledgehammer’ goes a different way. It will also offer 64-bit extensions, but it will still offer native support of 32-bit software as well.
There’s still only wildly speculative data about each processor, although Itanium is obviously much closer to its launch date. I prefer to not comment on any of those processors before we can spot some actual real-world data about each of them.