I have the same concerns regarding AMD K7 and its implications for Intel's stock price. Intel stock dominates my portfolio and I have analyzed this issue for the past few months . I have fair amount of expertise in using and optimizing x86 instruction code and I will give you my perspective.
AMD described the K-7 at the Microprocessor Design Forum a few months ago.
1) The K-7 has three Integer Execution Units as opposed to two on the Pentium II. Thus it will execute most integer instructions in 0.33 clocks as opposed to 0.5 on Pentium II . The K-7 has two Floating Point Units as opposed to 1 on the Pentium II . The K-7 will execute a typical floating instruction in 1.5 clocks as opposed to 3 clocks on the Pentium II. Thus the K-7 has the potential to outperform Pentium II at the same clock speed and without new software.
2) The K-7 has a larger L1 memory cache than the Pentium II and has added an L2 backside memory bus like Intel and increased the main memory speed to 200Mhz as opposed to 100Mhz on the existing Pentium II . The memory architecture is more critical than number of execution units for most applications but it is hard to quantify . The K-7 appeared to have a superior memory architecture at first glance . The importance of memory architecture can be understood by the following observations . If data is not found in L1 cache but is in L2 cache the processor is stalled for 1-5 clock cycles. If the data is not in L2 either the processor will stall for 40-70 clock cycles . [This number is not found in processor data sheet since it depends upon the peripheral chip sets and memory chips being used . This number is determined empirically]. Memory architecture controls computer performance.
3) AMD has to recreate the entire CPU motherboard infrastructure for the K-7 . Brand new processor cartridges, chipsets and mother boards have to be developed. AMD has no design expertise or manufacturing capability in these areas.[So far AMD has ridden Intel's coatails and used the infrastructure developed by Intel]. It will be an expensive undertaking and require lots of money. AMD is already committed to building a new Fab in Dresden which will eat up lots of capital.So either AMD will have to sell more stock and develop everything internally or depend upon other partners to build the infrastructure. The safe bet for AMD is to find other partners, and it appears to be working with Taiwanese chipset designers and motherboard makers. History is not on AMD's side. Intel depended upon other vendors for the 386 and 486 peripheral chips, and 386 and 486 computer systems were available 12 to 18 months after the 386 and 486 processor chips were available. Nextgen 586 was far superior to Intel 486, but it also required a new infrastructure and by the time this happened, Intel was well ahead of Nextgen. Problems with the K-7 infrastructure are surfacing already .It is rumored that the first K-7 chipsets have a L2 with 1/3 processor speed, and the main memory bus will only be capable of 100Mhz .If this is true K-7 cannot reach its potential. In addition the Intel infrastructure will be manufactured in huge volumes, and it appears that the K-7 infrastructure is doomed to be more expensive .So AMD's partners will tend to limit the investment. So the "AMD Hype Machine" will have to be continually in an "overdrive mode" to dangle the carrot before its partners.
4) Intel's Katmai Chip : Computer systems with Intel's Katmai chips will begin shipping Feb 28,1999.The Katmai will run at 15% higher clock rates than Pentium II for the same process. It will have new instructions which will execute 4 times as many Floating point operations[0.75clocks per typical Floating Point Calculation]. It has new instructions which can eliminate the memory loading penalties discussed in 2) .So on paper Katmai will have at least a 2x advantage over K-7. Unfortunately the software will have to be rewritten to unlock this potential. History is not on Intel's side on this issue . It has typically taken years before new software instructions are incorporated into common applications. Intel has been shipping Katmai demo machines to most Software Developers since June 98 . Intel claims to be making an extraordinary effort to have Katmai Applications ready by February. I believe that Intel may have compilers for Katmai instructions already. If the Software Developers have "Killer Applications" ready by March, or if Intel gives away a Katmai Compiler free to every Software Developer, then Katmai Compatibility will become a must, and K-7 will have to go back to the drawing board. [ "Intel Hype Machine" sold the world on MMX even though very little software takes advantage of it. MMX programmers have to code in Assembly Language, which is a lost art form .A MMX compiler would lead to widespread MMX usage ].
5) Coppermine and 0.18 micron Technology: At the Financial Analysts Conference Intel showed data which implied that the 0.18 micron process is ready to go into manufacturing. AMD has a lot further to go. AMD R&D personnel were shipped en masse to Texas to clean up the manufacturing mess at Fab 25. They succeeded in their assignment, but the 0.18 micron development suffered, and so AMD will be buying the 0.18 technology from Motorola. Tranfering technology from Motorola will not be easy, so Intel has a years lead on AMD in taking 0.18 micron technology to large volume production. Historical trends indicate that this should boost clock speeds by 50%. Thus even with existing software instructions K-7 may not have an advantage because most likely Intel chips will be running at a 50% higher clock speed .In addition Intel will be adding a large L2 cache on chip, further improving its performance.
In summary, AMD K-7 looks good on paper, but the complex new infrastructure will most likely delay the full realization of its potential .In the mean time Intel will raise the clock rate and use on chip L2 caches to boost the performance and blunt the potential advantages of K-7. If any Killer application uses the Katmai New Instructions, any K-7 advantages will be lost . So I don't think that K-7 will significantly impact Intel's profits . Intel is a hardened survivor of several processor architecture wars. Motorola 6800,National 3200,Motorola 68000,the “RISC bandwagon”, the PowerPC, the Nextgen 586 all had “superior architecture ” . Intel always leveraged its large installed base, its economies of scale and “exemplary execution” to defend its existing markets and extend its domination of CPU business. Intel was blind sided by the low cost PC, but turned on a dime and will regain its lost market share once the present capacity constraints loosen up.
Over the past 15 years Intel has been an outstanding investment, and I have continued to expanded my holdings .I do not think that the K-7 will make a significant dent in Intel stock price in the long run .However there may be short term peturbations caused by AMD's campaign to gain support for its processor.
|
