1999 in Review
The Disappointment of the Year???
i820. The chipset, which was half a year late, still failed to make up either for the time spent on useless waiting, or for the price we have to pay for the systems based on it. No matter, which combo you decide on: the price-to-performance ratio will still be the worst of all the today's available solutions.
Memory
The industry celebrated the coming of the New Year in an already common state: everybody was getting ready for the shift to a new memory type, but no one knew to which one. That's why just in case they prepared to as many situations as possible. Each large memory manufacturer worked in all three directions: PC133 SDRAM, DDR SDRAM, Direct Rambus DRAM. Exactly at that time Intel did its best to provide the acceptable amount of RDRAM chips by the time its i820 was expected to come out: its had just invested 500 million dollars into Micron, and then another 100 million into Samsung.
However, the computer industry was not very enthusiastic about this memory type. Most large manufacturers didn't plan to start mass production of RDRAM until the first half of the year 2000, although they were assembling new production lines and promised to start in summer 1999. Besides, to tell the truth, not all of them were equipped enough to manufacture 800MHz RDRAM chips, and as for 600MHz they could provide without any problems, it wasn't demanded by the customers. As a result of this disagreement, they came to a compromising solution: to manufacture a new Rambus version - 700MHz RDRAM.
By that time, SLDRAM developers (one of the three competing technologies besides DDR and Rambus) analyzed the situation and stopped working on this memory type. They announced absolute DDR support as well as their intention to totally focus on finishing DDR II specification. The corresponding chips were expected to appear in the beginning of 2001. In the end we got a classical scheme: DDR face to face with Rambus without any indefinite factors.
PC133 SDRAM, which non-buffered version had been already approved in the beginning of spring, couldn't be regarded as one of these disturbing factors, since it was not a candidate but simply a transitional stage on the way towards DDR. The spec is approved, and by the summer most leading memory manufacturers started mass production of this memory type. PC133 modules emerge in the price-lists especially since the first chipset supporting this memory type appeared right then.
And as for PC100 SDRAM, the situation still leaves much to be desired. Starting from the end of winter the companies got rather worried about the prices on PC100 chips, which dropped down to the manufacturing cost of the chips. However, this happened despite the fact that most manufacturers had just started to recover from the last year crisis. Some companies even failed to come round: Matsushita was forced to quit, Fujitsu gave to Acer its 64Mbit chips manufacture in February, closer to summer IBM announced that from then on their chips would be produced by Taiwanese Nanya.
Those who nevertheless decided to stay, of course, tried to reduce the manufacturing expenses on the memory chips, in other words they worked on improving the manufacturing technology changing it from the usual 0.25 to 0.20, 0.19 micron. Everything was balancing on the very edge, and finally Micron upset this unstable balance having started to empty its stores in a rather aggressive way and to reduce the prices. Surely Korean and Japanese manufacturers couldn't keep silent and responded to this gesture. So, in the beginning of July the price on 64Mbit chips dropped by almost two bucks during one month and reached the lowest level of less than $4 for a chip, while the manufacturing cost $5. So, this gave way to rumors about Micron's problems with the masks for 0.21 micron chips manufacturing, and then...
In the end of July the price rose again up to $6-6.5, then followed another few upsurges and drops, and in August everybody expected the price to get to 7 bucks. In fact, the price exceeded this number and reached $9.11. The only consolation, if it could only be considered a consolation, was the quick price reduction on 128Mbit chips, which began in mid July. And then? And then September set in. To be more exact the 21 September, when an awful earthquake shocked Taiwan. During one single month the price increased by over 100% - up to $21.09. The market was living through incredible times. Everything which could replace the unbelievably expensive 64Mbit PC100 chips simply disappeared from the shelves: 64Mbit PC66 got almost as expensive as its faster fellow, the price on 16 and 128Mbit memory chips also rose. And in October the events changed their fatal course and everything went vice versa: the price dropped almost as drastically as it had increased the previous month - down to $11.44. And then the process slowed down. So, by the end of the year 64Mbit PC100 chips went for 9 bucks each.
One of the consequences, which resulted from these stunts, turned the absence of any interest in RDRAM for the near future. There was simply no time for it since you could easily make money with the ordinary SDRAM, which manufacturing was already in full swing and didn't require any effort from you. Moreover, RDRAM kept surprising us with the whole bunch of various inconveniences: 32X testers of RDRAM chips were promised only in the first quarter, Intel postponed its i820 till September and then announced PC133 SDRAM support, provided really modest results shown by RDRAM and in the end postponed i820 for another indefinite time period.
After the last move like that, memory manufacturers decided to freeze RDRAM manufacturing until Intel solved all the problems. Two months later they resumed RDRAM manufacturing, but that autumn still told on its health condition. The companies didn't stop supporting RDRAM but significantly reduced the pace: they cancelled all the additional testers, equipment and other supplies, because they considered it cleverer to watch the market development first and then to make the final decision.
Rambus took a look at what was going on and expressed its desire to concentrate mainly on other market sectors, which required high throughput, which would do it justice and love and respect it. Besides, Rambus also announced in the meanwhile that they were going to increase RDRAM chips speed up to 1.6GHz. Memory manufacturers preferred to refrain from comments on technical aspects of this project. They only stated that speed increase wouldn't help solve the main problem of RDRAM - high cost.
A four-year collaboration between Intel and Rambus aimed at RDRAM promotion came to an end. Intel analyzed the results and the prospects and arrived at a conclusion that they had to admit their fault when they tried to impose RDRAM upon the computer industry. The company agreed with the five world's largest memory manufacturers upon the joint development of absolutely new memory architecture, as in good old times when Intel was not opposing the computer industry but guided its growth in this or that direction. This architecture allegedly coming after DDR II is supposed to accumulate all the best of both: traditional technologies and package RDRAM.
Unlike RDRAM, which struck us as too slow this autumn, DDR on the contrary started developing actively in the end of the year. Every manufacturer had already got 128Mbit 266MHz chip versions, and by the end of the year some of them shifted to 256Mbit chips. AMI2 announced that DDR modules reference designs were available for free. IBM opened a new section on their website offering free designs for modules and DDR based systems manufacturers. Hyundai and Micron kept promoting this memory type. And according to IDC, after Hyundai absorbed LG Semicon in October they became the first and Micron - the second world's largest memory manufacturers, while Samsung moved from the first to the third position. However, the situation may still change on 1 April 2000 when a new company comes into the world: NEC-Hitachi Memory Inc.
All in all, here the situation is absolutely different from what we saw with Rambus. The memory is available and all memory manufacturers are simply waiting for the starting signal. What's the problem then? No chipsets! And they will appear only in the first quarter of the year 2000, which will definitely speed up the market growth.
In the beginning of the year 2000 the server manufacturers will start using LE64 from RCC, and some other manufacturers will offer their server chipsets supporting DDR SDRAM. And Intel chipset will also be among those. However, the PC manufacturers won't fall into oblivion. Apollo PX266, AMD-760 and a new Intel chipset will be at their disposal here.
No one besides Intel announced their intention to develop chipsets with RDRAM support that's why here everything seems much more tranquil and indefinite. Mostly due to the fact that nobody knows and can foresee which direction Intel will choose, especially since it has already partially turned to PC133/DDR. In fact, the motion of most large companies is usually very much like the Brownian one.
RDRAM manufacturers are going to move in the only possible direction opposite to that of Rambus. They, the actual manufacturers, will have to limit Rambus engineers' imagination and to reduce the number of chip banks. Of course, it will result into slower memory, but on the other hand, not only the speed will go down but the price, too, which is pretty pleasing. Although we are not discussing the implementation fields of a cheaper and slower RDRAM while it will be possible to use cheap and fast DDR SDRAM. Maybe...
The Memory of the Year
PC100 SDRAM. Like 440X, which managed to remain the chipset of the year in 1999 as well, this memory type also retained its popularity and proved an indisputable leader of the year since there was simply no alternative to it at all for over 9 months. But this question should be addressed to chips manufacturers.
The Disappointment of the Year
Direct Rambus DRAM. Too expensive and at the same time too slow, this memory was simply knocked down by the constant problems with i820 and the manufacturers inability to provide mass amounts. What do we have in the end? Of the 200 million chips planned for the whole year the world hardly saw a couple percents.
ixbt-labs.com |
