Due to my recent visit to the Bay Area I could gather quite a bit of information about the future development on the CPU market. Here are some short summaries about what will happen next in the big x86 CPU companies.
My full day meeting with Intel in Santa Clara last week was a very pleasant experience. I wasn't only supplied with quite a bit of interesting information about various topics, but received also a very enjoyable and friendly treatment. Intel is certainly concerned about improving the relationship with the press and it was obious that they were choosing just the right people to accomplish this. Hence I'd like to announce my complete peace with Intel. I look forward to a friendly relationship in the future.
Before I will talk about Intel's latest CPU plans, I would like to recommend one thing for all of you that are interested in the actual CPU manufacturing process. If you should ever be in Silicon Valley, don't miss the Intel museum. There you can see all the various steps in chip manufacturing, starting at the production of a silicon ingot and finishing with the packaging of the chip. You'll find it in Santa Clara, Mission College Blvd., off Montague Express Way (reached from Freeway 101, 680 or 880). I only had about 15 minutes of rushing through it, but if you've got the time you can easily spend several hours in there.
Intel's spokesmen are extremely optimistic these days and I was wondering why. However if you hear the plans for the next 6 months you might understand why. First of all nobody is really talking of Socket 7 anymore, and if, then only in combination with MMX and notebooks. Intel will very soon release a .25 micron low power Pentium MMX, codename 'Tillamook', which will run at 2.5 V or even less, so that notebooks can take advantage of MMX applications in the near future.
The next thing, which I already announced about a month ago will be a Pentium II for Socket 8, called 'Pentium Pro Overdrive'. This CPU will come with something like a 'back package' where the external level 2 cache of the Pentium II will be located, combined with a heat sink. Since this CPU won't have much more in common with the Pentium Pro than the socket it plugs in, it will also reach the clock speeds of the Pentium II, which currently are 233, 266 and 300 MHz. Before ever getting this CPU please make sure that your board supports the higher multiplier settings of x3.5, x4 and x4.5.
The Pentium II is of course Intel's favourite child these days and hence you can expect the most news here. Within probably this year, but latest at the beginning of 1998 there will be 3 (three!) different kinds of Pentium II CPUs. One is the Pentium II we already know that fits into Slot One. The next one will be a smaller special low power version for notebooks, since you can imagine that the huge cartridge of the current Pentium II would hardly fit into any notebook, let alone the immense power consumption of this CPU. Number three will be the most interesting and maybe also upsetting version for all real power users. This third version will come with a larger level 2 cache that will run at clock speed instead of half the clock speed in the current Pentium II. Due to mechanical reasons (this is at least Intel's version) this new high performance Pentium II will come in a new and different package for a new slot that might be called 'Slot Two' (no confirmation from Intel about the name). This new high end Pentium II will also solve some maybe unknown limitations of the current Pentium II, which are
- only 512 MB cacheable area (Pentium Pro does 4 GB)
This seems to be a kinda serious problem, since the Pentium II would only run with more than 512 MB RAM in case you disable both caches, the L2 and the L1 cache, which decreases the performance of this CPU down to something in the area of a Pentium 100. Without disabling the caches, the system would crash.
- only running in dual CPU configuration, no quad or higher configuration possible (Pentium Pro can run at any multi CPU configuration)
Hence no quad-CPU servers possible with the current Pentium II.
Hence this new high performance Pentium II will first be targeted to the high end server and workstation market. Whoever wants to get this CPUs for his system will need a new motherboard with a new CPU connector, so be careful with buying a Slot One board unless you either want to stick to the current Pentium II or you don't care about buying another one pretty soon.
Shrinking the Pentium II die to a .25 micron size will soon raise the clock speed and lower the power consumption of new Pentium II CPUs. This .025 micron Pentium II is already known under the code name 'Deschutes'. The current CPUs with clock rates of 233 up to 300 MHz will continue using the .35 micron die. Intel seems positive to have a 333 MHz Deschutes out this year still and 366, 400, 433, 466 up to even 500 MHz in the first half of next year. I guess that's the reason why Intel spokesmen are so optimistic, because since the Pentium II is scaling so nicely over the clock speed (due to the level 2 cache running at 1/2 or 1 times the clock speed) the performance levels of a 500 MHz high end Pentium II will give Intel's competitors quite a bit of a head ache. The Deschutes will come in two different flavours, first the 'mainstream' version that is also limited to 512 MB of main memory, later on a special server version that will be able to cache up to 2 GB.
The high end Pentium II for 'the other slot' will be designed to run at 100 MHz bus speed in the first place, although there will be the chance to run it at 66 MHz bus as well.
One little word to the next Intel Pentium II / Pentium Pro chipset, the 440LX. You probably know that the most important new features of this chipset will be SDRAM support, AGP support and Ultra DMA support. This chipset will be launched pretty soon on August 26, 1998. Intel isn't very happy however, because Microsoft is pretty late with its release of 'Memphis' (Windows 98) and Windows NT 5.0, which will be the first operating systems to support AGP. There seems to be a simular problem with Microsoft's I2O support in their OS versions, so that Intel currently has enough reasons to be a little moody with Microsoft. This is also giving problems to me as a hardware tester, since I need an AGP enabled OS for my AGP testing in the very near future. I have to stick to beta versions, which obviously don't necessarily show the whole performance advantage. Some AGP testing I did a few weeks ago couldn't show any enhancement from AGP at all, due to alpha drivers of the cards and other software odds as well as a 440LX chipset revision level of only 'A1'. The stepping 'A2' is out now and the next testing will hopefully be more revealing. I'll keep you posted.
The following chipsets will be the 440BX and 440NX. Both are supposed to run at 100 MHz bus clock, using SDRAM. The 440NX chipset will be for servers and will replace the 450GX chipset. It will support quad or more CPUs and will hence run with the new high end Pentium II.
One last word about nDRAM, Intel's future RAM technology, developed together with and basing on Rambus RAM. This RAM will probably come out together with the 'Merced', Intel's future 64bit CPU, developed in cooperation with Hewlett Packard, that'll run at up to 1GHz. Hence no nDRAM much before the end of the century.
Although I've got a very friendly contact with AMD, news about a future CPU are currently zero. However I at least don't get my bag searched when entering or leaving their building as happening at Intel in Santa Clara. The news I do have however aren't too bad as well. You might know about VIA's upcoming new chipset called Apollo VP3, which will include AGP support for Socket 7. AMD is also working on their own AMD640AGP chipset, which doesn't seem much behind VIA. Hence we shouldn't have to wait too long after August 26 to welcome AGP to Socket 7 boards. Boards with the AMD640 chipset are starting to ship very soon too, I've already seen one so far, but I'm not yet allowed to publish about it.
AMD's fab in Austin seems to work fine now, so that AMD will eventually be able satisfying the market with their K6 CPUs. They are also working very hard on the first shrink to .25 micron, which will lower the voltage and hence the power consumption of the K6 as well as finally enable the manufacturing of 266 MHz or even faster K6 CPUs.
One very interesting thing got almost forgotten in the last months, which is AMD's plans of the high performance 100 MHz bus. From the information that I've got it looks as if AMD will release this new technology long before Intel (440BX and 440NX, both Q1 1998). I look forward to testing a 100 MHz bus speed board pretty soon.
Although Intel's plans sound pretty scary, AMD doesn't seem to be impressed at all. I'm pretty sure that AMD has got an answer to the 500 MHz Pentium II, however so far they are keeping extremely quiet.
I didn't publish the bad news I heard from IBM about 4 weeks ago, because I didn't really want to believe it. Now it seems as if I've been right about it, since things are looking better for the 6x86MX now. These 4 weeks ago I was told that due to manufacturing reasons Cyrix and IBM would only ship 6x86MX CPUs which run at x2 multiplier, 133/66 and 150/75 MHz CPUs, which weren't even announced in the first place. IBM wants me to point out that these were pure yield problems and didn't have anythin g to due with flaws in the manufacturing process. Now it looks as if the yields for the faster and real ones are improving, so that you soon should be able to at least get a PR2 166 (150/60 MHz) and a PR2 200 (166/66 MHz). The PR2 233 (187/75 MHz) will hopefully be in mass production by September, faster clock speeds will follow in Q4 this year.
The currently shipped version of the 6x86MX is revision 1.3.5 and Cyrix/IBM are now testing the new upcoming revision 1.4, which will already be the first shrink. This revision (1.4) should be able to push the yields towards 187 MHz and above.
There also seem some logistic problems that caused IBM being a little bit behind Cyrix in terms of available 6x86MX CPUs. However IBM is ramping up production of their own CPUs as well now. Cyrix asked me to make clear that there is no difference whatsoever in the validation process between Cyrix or IBM CPUs.
The voltage requirements of the 6x86MX have changed from 2.8 V to now 2.9 V. This is valid for Cyrix as well as for IBM.
Cyrix 6x86MX CPUs doesn't show any marking that would tell you the revision number of the chip and they don't seem to be happy about IBM having this 'revision level' marking on their 6x86MX CPUs, which is the last letter in the OEM product number printed on the chip (currently 'A' = revision 1.3). To stay conform with Cyrix, IBM is now considering of not making the revision level coding publicy available anymore as well, although you will still able to see that the revision number must have changed when the 'revision level' letter has changed. In both CPUs you will still be able to read the revision number out via the CPUID registers. However do I have a problem appreciating Cyrix' policy of keeping us more in the dark about their CPU revisions. What do you think?
Here the IBM 6x86MX spec number from the data book:
Since the Computex I haven't heard much of the C6 anymore, which sounds as if they've got problems getting up to 200 MHz versions. These 200 MHz C6 CPUs were promised to several press people including me as soon as they are ready. Obviously they aren't ready yet, which will mean that the end user will have to wait even longer until this CPU will be avilable in the shops.