by FastSite
01/19/2000 | 12:00 AM
Cyrix processors have always been known for their super low cost. At the same time AMD tried to remain a typical middle class, and Intel traditionally skimmed the cream off. And in fact, the financial state of these companies by the end of 1990s reflected the same tendency. Besides, there was one more thing that had a very serious influence: the trend towards sharp PC cost reduction, which began in the last couple of years. However, it didn't turn Cyrix into a King, as you have probably expected. On the contrary, it attracted Intel and AMD's attention, so that they have been fighting violently in a price war for the last two years (which led to the same destruction as an elephant could cause to a china-shop :-).<%BANNER[article]%>
And the result is well-known to all of us. In summer after a relatively long lasting agony of its MII family National Semiconductor had to leave x86 processor market, which was evident in the end of 1998 already. However, they were not the only ones who were forced to quit. A couple of other companies working in the same market sector also retired. These were IDT and Rise. But their leaving in no way meant, that the demand for $20-30 processors suddenly disappeared. On the contrary, it remained, and maybe even rose a bit higher since the requirements to office software packages stopped growing in the last years, so that the performance of the CPUs of this kind proved more than enough to satisfy all the office needs. So, the demand remained, and the supply almost vanished. Well, this situation needed immediate improvement.
At this hard time VIA Technology came to rescue. By the beginning of last year their ambitions grew up to incredible highs. This company has been very successfully competing with Intel in the field of chipsets recently, which was mostly due to its brilliant skill of selling cheap products. And this time they decided to apply the same approach to processors. Buying another company, which had already achieved something in this field seemed the easiest way here. Moreover, since fortune turned its back to VIA and the latter at any rate needed to get access to patents, which had something to do with GTL+. This summer VIA reached its aim having purchased two processor developers groups - Cyrix from National Semiconductor and Centaur from IDT. And although Centaur became VIA's indisputable favorite, the first processor they launched would still be a one from Cyrix, which is actually the subject of our preview.
The first information about this product appeared in autumn 1997 at MicroProcessor Forum, which had by that time become a traditional occasion for the announcements like that. Cyrix vice-president Robert Maher described its Cayenne core as a modified core of 6x86 family supporting MMXFP instructions, which is a set of SIMD instructions developed by Cyrix. The integer unit remained almost untouched, while the floating point unit had to undergo some considerable changes. There appeared pipelining and the possibility to operate two floating point numbers per time step. The new coprocessor had to raise Cyrix processor practically up to the level of Pentium II/Pentium III. The same thing can refer to the way Cayenne worked with MMX instructions.
As for the technical parameters of the new processors, they implied that the manufacturing of Cayenne would start in the second half of 1998. It was supposed to be made with 0.25 micron technology (with the chip surface size of 70sq.mm composed of 6.8 million transistors). Within 1998 its working frequency was expected to grow from 250 to 350MHz (PR rating - from 300 to 400MHz). And in 1999 a cooler Jalapeno had to replace it.
Nevertheless, 1998 was over, and the first half of 1999 was over, too. Cyrix had already managed to renounce their own instructions set in favor of 3DNow!, however Cayenne was not yet ready and the developers kept working on its improvements and enhancements. Besides, it changed its name to Jedi and then to Gobi forced by numerous claims from LucasFilm. As for the design changes, it acquired a 256KB L2 on-chip cache, as it had been predicted at MPF'97. The cache was of the same size as that by today's Coppermine. Besides, you should also try to recall 64KB single L1 cache, which was twice as big as that by Coppermine. This was how the things stood at the moment when VIA purchased Cyrix.
By that moment the processor was ready to be born: some time before it had already been demonstrated at Computex'99. So, having bought Cyrix, VIA not only got the patents it needed, but also a relatively cheap but rather efficient and fast processor, which had undergone a two-year improvement. So, although Cyrix dismissed about fifty percent of all its employees, VIA insisted on leaving about 150 people, so that within a few more months they could complete the processor and make it ready for commercial manufacturing.
By that time the processor started looking as an almost finished product: the interface was finally defined as Socket370, the system bus frequency - as 66/100/133MHz, the manufacturing technology - as 0.18 micron. But the most important thing, which had been finally settled, was the name - Joshua. In fact, the most important thing is of course the fastness, but here nothing was clear yet. When we were writing this preview, we supposed that the PR-rating of the processor would be 433/466/500/533/566MHz. However, by the deadline only the first two values of the family are likely to be introduced.
As for the overclockability, the answer is rather vague here. On the one hand, the clock multiplier isn't locked, which seems heavenly cool compared to Intel and AMD processors. On the other hand, the processor initially designed for the voltage of 2.2V (for the slowest 300MHz model marked as 433 according to the PR-rating). However, the chosen technology - 0.18 micron - still gives us hope for the better.
As for the performance, we expect it to be at the level of Celeron processor with the working frequency according to PR-rating. Some drawbacks of a slightly improved architecture should be compensated mostly by 256KB L2 cache working at the processor clock frequency. In general, it has to do with office applications, where the extra 128KB L2 will hardly be a burden. Speaking about the games, we have to mention a significantly improved coprocessor since the times of 6x86MX. And of course, you shouldn't disregard the support of 3DNow!. Today almost all the more or less serious games as well as new graphics card drivers are optimized for this set of SIMD instructions.
All these things will be available for a very low price (our reliable sources in VIA Company particularly stress this phrase). Taking into account that according to Intel's recent info, its Celerons won't keep up with the same fastness and their frequency will hardly exceed 600MHz. But as for VIA, they seem very self-confident and not for nothing, actually.
Anyway, it is the price that matters in the Low-End market (and devotion to the brand is of no great importance here). If VIA manages to take advantage of the PR-rating and to convince its customers that its processors go side by side with Intel Celeron though for the price twice as low as that of Celeron then Intel and AMD have every right to start feeling nervous about their Celeron and Spitfire.