VIA Apollo Pro 133A & VCSDRAM - Part 1: The Technology

One of the features of both the Apollo Pro 133 and the new 133A is the support for a technology known as Virtual Channel Memory, often abbreviated as VCM, VCSDRAM, or VC133. With the industry seemingly unhappy with the way Rambus is performing, especially for the price, the search for a true competitor is underway.

The competitor everyone is looking for is one that can boast the low latency of SDRAM with the higher bandwidth that RDRAM offers. That competitor doesn't seem to exist outside of vaporware. The ideal solution seems to be Double Data Rate SDRAM, or DDR-SDRAM. DDR-SDRAM offers the low latency of SDRAM while effectively doubling the amount of available bandwidth by transferring on both the rising and falling edges of the clock (ala AGP).

With PC133 SDRAM offering around 1.06GB/s of bandwidth, DDR PC133 or PC266 SDRAM would offer 2.1GB/s of bandwidth, that's 30% greater than PC800 RDRAM currently offers. Like RDRAM, DDR-SDRAM has been "around" for a while. For example, RDRAM was even used on some older graphics cards as well as on Nintendo's N64 gaming console, but it is just recently making its way over to the PC. The same can be said about DDR-SDRAM, but in a different manner. DDR-SDRAM has been around for quite some time but only in theory. Actually producing DDR-SDRAM chips hasn't occurred in respectable quantities until very recently. In that respect, DDR-SDRAM is still a sacred stone in that it would solve our bandwidth problems but no one can seem to produce high enough quantities of it to push the standard forward.

In the meantime, NEC's solution to the problem was with the development of Virtual Channel SDRAM, a technology designed to improve the efficiency of SDRAM and thus increase available bandwidth to the applications that need it. How does this work? Let's take a look at NEC's approach to the bandwidth problem…

The first concept that you need to understand is the concept of memory masters. A memory master is anything that accesses the main system memory. Your CPU, your AGP card, and your PCI devices (actually the PCI bus itself) are all memory masters. What do all of these have in common? They all try and access system memory, often times, concurrently.

Now, each one of these memory masters has one particular goal in mind -- to get what it needs from the system memory. Your AGP card doesn't care if your PCI cards needs to retrieve some data from memory; it only cares about the fact that there are textures in memory that it needs to retrieve so you can play your games. At the same time, your SCSI card doesn't really care that the AGP card needs to retrieve its textures, it cares about its needs for memory. As you might be able to guess, this can result in inefficient usage of memory bandwidth, especially if multiple memory masters are trying to access data stored in system memory at the same time.