Memoirs and chipsets at the beginning of 2000
The chipset i820 and his problems of memory
Even his most fierce enemies admit that up to the date Intel has made some of the best chipsets that we have seen, what does not turn out to be surprising if we bear in mind that most of the microprocessors and technologies to which they had to do these chipsets to work were made by the same company.
Nevertheless, at the time of designing 820, Intel committed what is demonstrating to be a big forecast mistake: he designed it to work exclusively with memory RDRAM (Rambus DRAM), thinking undoubtedly about the good yield of this memory, but very probably also like a strategic decision of market, since Intel has invested a lot of money in the company Rambus and in the proper RDRAM.
Nevertheless, since we have seen previously, the Rambus has demonstrated to be a memory very difficult to make and disproportionately face, for what Intel turned out to be forced to include an additional chip (called MTH, Memory Translator Hub) to offer also support of memory SDRAM (PC100/PC133) in the basic badges with chipset 820. This MTH works of "translator" between the SDRAM and the chipset 820 (which really continues without supporting the SDRAM).
This translation process not only gets dearer the price of the badges (more chips and connections = more expensive; some question?), but also it provokes that the yield of a badge with chipset 820 and memory SDRAM is even worse than of one with chipset BX. Namely that we can be condition on having updated our "old" badge BX to the more modern one (and face) 820, the yield goes down. The world the other way round...
And this is not quite; to finish off the task, the MTH does not support SDRAM PC133, but only PC100. Very well, it is not that we could not install memory PC133, but if we it do the MTH it will make her work to 100 MHz, therefore well we can save ourselves the difference and buy straight PC100...
... Of course, fortunately the fact that the memory works to 100 MHz he does not prevent us to install mikes with 133 MHz bus (in these badges the buses of memory and of system or mike work of independent form), but it is not a big consolation, the truth.
If these problems were small, even in case he could pay the Rambus, at present the maximum number of grooves for memory Rambus is 3, and many basic badges only they have 2 grooves. This owes to a mistake in the proper chipset 820 to itself that Intel has been unable to correct on time, and implies that ideally the badges are seen limited to 768 ó 512 MB (with 256 MB modules ). If we use 64 MB modules , the maximum limits to 196 ó 128 MB, what at present is a too exact quantity.
Being just persons, we must admit that yes there exists a chipset Intel who supports up to 2 GB of RDRAM in up to 8 grooves, the Intel 840; but after this chipset destined for servants is, the basic badges that incorporate it cost a minimum of 50.000 ptas, more enough than an average user would consider reasonably.
PC133, Apollo Pro 133, BX and KX133
If the Rambus seems reserved to the one who could pay it: what solutions do we have the mortal poor poor? To use memory SDRAM (PC133 or PC100) and the chipsets ROUTE Apollo Pro 133 or Intel BX.
The best option seems to be Apollo Pro 133, one chipset very modern that in his 133rd version it includes up to support AGP 4x (the normal version includes AGP 2x, although it is not a very significant difference). He does not support Rambus, but SDRAM (of any speed) and a variant of the same so called "VC-DRAM" that increases lightly the yield.
Otherwise, he supports all the current technologies: 133 MHz bus , UltraDMA66, AGP 2x ó 4x, up to 1,5 memory GB (if the badge has enough grooves) and even 4 ports USB. The only thing that begins in absence support multiprocessor is, therefore dual badges do not exist with this chipset; but really immense most of users do not use this capacity, and if he wants to mount a servant already has 840, the BX or the GX.
Despite his seniority, the BX can still give many war. True it is that officially he does not support 133 MHz bus (after 1/2 have no divisor for the AGP, we would be overclockeando the graphic card), he does not support memory Rambus and his maximum speed AGP is 2x; but many exist Pentium III with 100 MHz bus (both with ancient nucleus and with nucleus Coppermine "E"), he supports up to 1 memory GB, the difference between AGP 4x and 2x is practically invaluable, it has support multiprocessor... and the most important thing, it is a placed, trustworthy technology, with many basic badges at very good price and an exceptional yield (if we use mikes with 100 MHz bus , superior to that of an Apollo Pro 133).
Let's not forget the chipsets for the one that many consider to be the most rapid microprocessor of the moment, the AMD Athlon K7. Till not long ago the chipsets for this mike only were supporting memory PC100 (although, as we have said, normally there can settle memory PC133 although it will work only to 100 MHz), but this situation has changed with the appearance of the excellent chipset ROUTE KX133, with support for PC133, AGP 4x, UltraDMA66 and 4 ports USB. It seems that the errors of Intel and the support demonstrated to the microprocessors AMD are giving an important impulse to her in other times almost unknown ROUTE.