The graphics pipeline and memory system bus have roughly doubled in their respective data handling capacities. The ICH, or I/O Controller Hub, has also incurred a doubling of maximum data moving across its bus, from 133MB/sec to 266MB/sec. This helps provide one of the strongest advantages for i820 versus 440BX, which is the separation of the ATA-66 channels off of the main system bus.
By dedicating a certain area off of the main system bus for the IDE channels, the i820 provides ATA-66 support for disk activity with virtually no performance hit to the main system. We'll talk more about this achievement in real world subjective results later in this article.
None of the improvements listed above would be possible without the massive bandwidth improvements that Intel has selected RDRAM for, so let's move on to this controversial topic.
Ever since Intel decided in the mid 90's that the memory technology developed by Rambus Inc. was critical to their progress towards improving their PC's bandwidth, the controversy over this decision hasn't let up.
Even today, roughly seven years after Rambus Inc. introduced the foundation of their technologies and patents to the industry, there is still controversy surrounding three main factors:
Cost
Performance
Availability
The reason for RDRAM's cost controversy is obvious. In an age where sub-$1,000 PCs are now the norm, no one in the industry wants to see memory prices higher than a dollar a megabyte, let alone the five to seven dollars per megabyte that PC800 RDRAM is expected to cost this year.
The controversy surrounding RDRAM's performance was originally started by Intel's partners and their competitors alike, both of whom sought to discredit the proprietary technology for several reasons. This is a complex issue, and it needs more explanation for readers to understand why credible sources in the manufacturing side of the industry have issued public statements decrying the new Intel technology over the past three years.
In the beginning, no memory manufacturer favorably looked at RDRAM as an economically viable alternative to the easier and cheaper to manufacture SDRAM which they had tooled up to produce in bulk by late 1997. The cost associated with creating the .18 micron RDRAM RIMMS along with the royalties that the manufacturer would have to pay directly to Intel for each licensed RIMM sold was undesirable to an industry that was used to setting its own pricing and determining the schedule of new technology implementation.