|Page 1 of 7|
|Author: BK Toh|
Date of review: 12-August-2000
Type Of Review: Mainboards
1. Introduction: A History Lesson
A year ago, selecting a motherboard for an Intel CPU was pretty straightforward. Everyone bought a BX chipset motherboard. Period.
All you had to decide was what features you required, and even then, the feature set was quite straightforward, with only ABit differentiating themselves by offering SoftMenu BIOS setup which gave the overclockers a huge boost. But this gap was quickly narrowed when other manufacturers soon wised up and added the same features.
Every manufacturer sold a motherboard based on the Intel BX chipset, and there were no viable offerings from ALi, VIA and the other chipset manufacturers, who could only supply chipsets for the lower end of the market (Pentium MMX and K6-2/3 CPUs)
As time went by, the manufacturers managed to improve the performance of the basic BX chipset and also sought to incorporate additional circuitry to keep the BX chipset on par with prevailing technology, such as supporting Ultra DMA66 hard drives by means of third party UDMA66 controllers from Promise and HighPoint, among others.
The BX chipset was initially designed to work with both 66MHz and 100MHz Front Side Bus (FSB) CPUs such as the Celeron and Pentium II CPUs. With the latter generation of Pentium III CPUs, an FSB of 133MHz was required. While motherboard manufacturers have always allowed for an overclocker to tweak the FSB of a BX board higher than 100MHz and beyond, this required some of the components (primarily the AGP bus) to run off specification, and could seriously create problems with stability.
When the time came for the BX to be phased out, however, Intel made a marketing error, and immediately the floodgates opened.
The successor to the BX, the Camimo 820 was a very flawed chipset. There were several things wrong with the 820 chipset. First of all, it depended on a dubious RAM technology called RAMBUS. While RAMBUS was rated at impressive speeds like 600-800MHz, the effective bandwidth was not very substantial, since it operated a 16-bit bus. This translated to roughly to a bandwidth of 1.2-1.6Mbytes/sec. With PC133 SDRAM operating at 64bit, the bandwidth was effectively 1.0Mbytes/sec. However, RAMBUS also required higher latencies than traditional SDRAM, so any improvements in memory benchmarks were marginal at best. It also did not help that there were manufacturing delays. When the first 133MHz FSB Pentium III CPUs arrived, Intel’s own chipset wasn’t even ready!
Secondly, the licensing agreements required RAM manufacturers to pay RAMBUS (the company) to manufacture RAMBUS memory modules resulting in highly exorbitant prices! I recall that at its peak, it was roughly about 10x the price of regular PC100 SDRAM modules.
According to Tom (of Tom’s Hardware), Intel had a strategic stake in RAMBUS and perhaps, the decision to go with RAMBUS had more to do with economic sense than technological considerations.
To soothe the budget conscious buying public, Intel hacked on support for PC100 SDRAM and called the new product CC820. In essence, Intel cobbled on some extra circuits, to translate RAMBUS signals to PC100 compatible ones. But the final product was stillborn. It offered dismal performance, especially when compared to the old BX chipset. In fact, even PC800MHz rated RDRAM had problems keeping on par with the trusty old BX. More recently, Intel faced another marketing nightmare when it was discovered that the CC820 had compatibility problems causing the computer to freeze or lock up intermittently, and Intel had to recall the motherboards.
In any case, the public made up their minds and voted with their wallets.
It was in this environment that VIA snapped the baton from Intel. When the 133MHz FSB Pentium IIIs (PIII-B) first arrived, VIA was first on the scene with chipset that officially supported the 133MHz FSB, called the VIA Apollo 133. Intel was nowhere to be seen since there were some last minute delays in the 820 chipset.
While the performance wasn’t as good as the BX, the VIA 133 nevertheless provided official support for the 133MHz FSB. Even respected manufacturers like Micron began to ship their first PIII-B PCs with a VIA chipset.
While Intel was still recovering from that blow, VIA delivered their second WHAMMY: the VIA Apollo 133A. Recognising the faults of their first offering, VIA went back and tweaked the design of the chipset and offered improved performance, rivalling the BX in most respects. In memory benchmarks, the 133A, like its predecessor, fell short of the BX, but this was a reasonable compromise since the BX could not officially support 133MHz FSBs. VIA also took the opportunity to provide AGP4x support and improved UDMA66 support.
Eventually, Intel had to go back to the drawing board and only recently delivered the 815E (dubbed Solano) chipset, which offers true PC133 SDRAM support, AGP Pro (to allow for higher powered graphics card in the future) and UltraDMA100 hard disk support.
In the meantime, users became increasingly aware of the specifications of the VIA Apollo 133A and were able to tweak the memory interleaving technique substantially, and could now offer BX-level memory performance. VIA Hardware has written up an exhaustive guide on how you can configure your Apollo 133A motherboard if your motherboard does not allow you to change the interleaving settings (which is disabled by default) in the BIOS. Thankfully, my VIA 133A BIOS provides the interleaving settings so my job was that much easier.