Hard Times At Intel High:
If you have even a remote interest in computers and the technology that drives them, you are pretty much familiar with who Intel is and what they have done for the personal computer. Chances are you have worked on a system sporting a 8086, 286, 486, Pentium, Pentium II, or Pentium III system during the course of your life. It simply can't be helped. You can't blame Intel, as the demand for their chips made them the dominant supplier of processors and technology for the computer industry. New competition would surface, but they could not keep up with Intel's marketing clout and superior technology, which until recently made alternative processors slower than an equal Intel counterpart. (Of course, that changed with the introduction of AMD's Athlon processor, but more on that later).
With a combination of strong leadership and leading edge technology, people were hard-pressed to find a better solution for their needs than an Intel processor. During those early years, Intel remained strong in their commitment to make technological advancements that kept them in the forefront of the technology sector. But something began to happen to mighty Intel.
Years of being at the top with no challenger that could even come close to touching them made them lower their guard. They started to care less for the consumer and instead focused on locking down the microprocessor market. Of course, this meant that their new processors contained less and less innovation. Subsequent processor upgrades where accompanied only by increases in the clock frequency, while containing little in the way of extraordinary new features.
So how did Intel go from being the leader of an industry that everybody was trying to catch up to, to finding themselves lagging behind one of their fiercest rivals? Lots of bad luck, poor judgment, product development problems, and for the first time, a company that actually had better technology than Intel.
The problems started some time ago, as Intel was making plans to unveil a new chipset, the i820, that would replace the venerable 440BX chipset that has become the standard of reliability and compatibility in the PC motherboard market. They would also soon be releasing a new line of microprocessors that would work with this new chipset and at a higher Front Side Bus than the BX could support, 133Mhz. But, Intel made four fatal mistakes in following their plans. First, they were pushing a new DRAM memory architecture that would go hand-in-hand with their new chipset, known as RAMBUS. RAMBUS is an IP (Intelectual Property) company that, with the financial backing of Intel, developed a new, high-bandwidth memory architecture that would deliver huge performance increases in overall system bandwidth as opposed to systems using today's standard SDRAM memory. At least that was the theory...
At the same time, chipset manufacturer rival VIA Technologies was pushing a new SDRAM memory standard that would replace the PC100 spec. This new spec, called PC133, called for SDRAM memory modules to work reliably with the new processors that ran at 133Mhz Front Side Bus, as well as adding support for the upcoming AGP4X spec. Most of the major system integrators and memory manufacturers agreed with VIA, for the simple fact that licensing RAMBUS technology was really expensive, not to mention that RAMBUS memory modules were at least three times more expensive than their SDRAM counterparts.. This prompted VIA and its backers to ask Intel to include PC133 support on their new i820 chipset. At the time Intel was not having any of that, since they had already spent too much money on RAMBUS. And here they make their second mistake. Including support for PC133 SDRAM in their i820 chipset would slow user's conversion from PC100 SDRAM into RAMBUS memory. Little did Intel know they would regret not participating in shaping up the PC133 spec.
The third mistake that practically sealed the fate of RAMBUS memory and the i820 chipset were the constant delays and technical difficulties encountered by Intel's engineering team in getting i820 to work reliably with RAMBUS modules. Problems included system instability when more than two RIMM slots were occupied with RAMBUS memory. This meant that the first run of i820 motherboards that contained three RIMM slots could not be used to their full potential. Then there was the fact that early benchmarks showed that an i820 plus RAMBUS system did not show a significant performance edge over an equally configured 440BX plus SDRAM system. These delays and difficulties kept pushing back the release of the chipset, which RAMBUS required in order to break into the market. Many of Intel's top customers, like Dell, Gateway, and Compaq, had to keep postponing the launch of their new systems since the chipset that went along with them were not ready.
Meanwhile, VIA Technologies was busy pushing their new Apollo Pro 133 chipset, which included support for both PC100 and PC133 memory, as well as support for the 133Mhz Front Side Bus. Later, the 133A revision added AGP4X support, as well as added speed and stability. This lead top system manufacturers to drop Intel's i820 chipset and RAMBUS in order to jump into the VIA bandwagon...at least until Intel got their act together. Of course, Intel did not take that lying down, and took VIA to court citing several copyright infringements. Either way, the Apollo Pro quickly made a niche for itself, and Intel needed an answer...fast.
Still not wanting to cave into supporting VIA's PC133 spec, (something all other manufacturers were clamoring for anyway), Intel makes a compromise. Intel agreed to release i820 with SDRAM as well as with its native RAMBUS support. How do they accomplish this? By adding a memory translation hub design. This hub translates RAMBUS memory calls into signals SDRAM can understand, which then allows us to upgrade to i820 while still using our tried and true SDRAM modules. The problem is that using the MTH penalizes you with a huge performance drop. The drop was so great, that 440BX systems beat out i820 plus SDRAM systems by wide margins in the most popular benchmarks (6-16% slower than BX chipsets). Users then questioned why they should upgrade to something that is slower than their current setups. They also questioned upgrading to RAMBUS, since the prices for the memory are astronomical, to say the least.
As if these problems were not enough, AMD successfully introduced their new line of Athlon microprocessors. The press quickly snaped them up and started to benchmark them. They quickly noticed that the new processor consistently beat similarly configured Pentium III Katmai processors. The best part is that Athlon processors are generally cheaper than Intel's speed counterpart. The pressure of seeing top system manufacturers adopt AMD's new baby probably prompted Intel to make its last mistake.
Intel was working on their new 133Mhz processors that would debut with their new chipset and memory technology, codenamed Coppermine. These processors were built on a smaller .18 micron size, and also required less energy to run, which kept them pretty cool. These new processors would also come in FC-PGA and SECC2 formats, with the FC-PGA processors debuting first. Intel even matched AMD's processor speed ramping, matching up quite well. But, the new processors were not quite ready for prime time.
The early Coppermine FC-PGA production runs yielded chips that had flaws in them. Intel calls them errata, but we generally call them bugs. The most damaging one to Intel's reputation and that of the new chips was the fact that on certain occasions, systems would fail to boot when first powered up, requiring a second push of the power button to make them react. Intel's PR quickly went into damage control mode, appeasing the worries of its clients by stating that only 1% to 2% of the Coppermines produced up to that point were likely to show the flaw. When that bug was discovered, the list of erratum found on the Coppermine reached 40 items. Since then, that number has risen up to 60 items.
In the rush to try and take some of the air out of AMD's launch, Intel rushed the introduction of what would be their next-generation computing platform, which included the i820 chipset, the new Coppermine processors, and RAMBUS memory. Quality control had sunk to an all-time low, allowing a processor with that many problems to make it into production. Granted, most users won't be afflicted by these problems, but still, the problems are there. In fact, Intel has taken this problem so seriously, that it has decided to make a new processor stepping in order to correct all of these flaws, taking the Coppermine from an A2 stepping to a B0 stepping, starting on April 7th. Still, users won't see these new processors until Intel's current stock is sold.
As if that was not bad enough, Intel kept announcing new Coppermine processors that were running at or higher than AMD's Athlons to steal some thunder from AMD's announcements. The problem was that Intel, even though they announced the chips, would not actually start shipping them until months later, (case in point, the 800Mhz Coppermine). AMD could afford to announce new processor speed stepping with ease, as their yields (previously their Achilles heel) were very high, and their new Dresden fab was starting to produce full-swing.
The latest setback to hit Intel happens to be that all February shipments of their 550Mhz Coppermine FC-PGA processors have been cancelled after the contractor that supplies the chip's flip-chip casings made a mistake. After that, top computer system vendors Gateway and Dell, blamed their low 4Q earnings to Intel's inability to supply enough of the new Coppermine chips to keep up with customer demand. This has prompted them to reallocate available chips to their most important customers.
Intel has changed its stance many times since then. For example, they will release a new chipset, i815, codenamed 'Solano', that will include native SDRAM support, eliminating the embarrassing performance hit i820 users experienced with the Memory Translation Hub. Probably the biggest shocker came when Intel announced it had partnered with the top five DRAM manufacturers, such as Hyundai, Micron, Infineon, NEC, and Samsung, to define and implement a new high-speed DRAM architecture for servers and workstations. Besides loosing some faith in RAMBUS, there was another reason for this move, and that was that Intel did not want to be left out again from defining new memory architectures. They learned that lesson after letting VIA and its partners implement PC133 memory without their input, (because they didn't want to). Just like that, Intel has begun the funeral proceedings for RAMBUS memory. Oddly enough, RAMBUS was invited to join this group, but it has not done so yet. Mainly because the memory manufacturers are pushing for an open architecture standard, riding themselves of licensing royalties, which is RAMBUS' business model.
So what has Intel done to get back on the horse? Well, besides the already mentioned plans to make a new Coppermine revision, they will also complete and start operating a new 300mm wafer fab in Arizona, which should help them achieve greater yields than ever. Finally adding native support for SDRAM memory was also a wise idea.
What the future holds for the i820 and RAMBUS memory is uncertain at this point, though many industry analysts do agree that their outlook is not very good. Some are predicting that DDR RAM and the i815 chipset will overtake and finally subdue the i820-RAMBUS combination. But until that happens, RAMBUS and i820 remain the top performance ticket in town, but only by a slim margin.
Never in my many years of watching and reading about this industry, have I seen Intel commit so many mistakes and so close to each other. If you believe in Murphy's Law, you could argue that Intel had it coming. After all, no company can ever stay at the top forever. I guess we will have to wait and see what Intel can come up with now. One thing is for certain though... Intel won't be held long as a follower. Look for them to come stronger than ever this year. And you all know what that means.. :) Us consumers get all the benefits.. :)
agnhardware.com
Sorry if duplicate post.. Don't have time to read 200+ posts tonight.
Milo |
