POST OF THE DAY
Rambus
Stifling Innovation

Format for Printing

Format for printing

Request Reprints

Reuse/Reprint

By aschwebel
December 20, 2000

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

With the two antitrust lawsuits sitting in courts, I wanted to research this a little bit. The whole notion about an antitrust case is that a company is both stifling innovation and hurting the consumer. Well, Micron and Hyundai have both taken this track in trying to fend off Rambus' royalty claims. What is their claim? Let's take a look at the landscape since 1993 when the Intel Pentium Processor was released.

Specs and Conclusions

Intel Processors

  • Pentium in 1993; Frequencies (60-200); FSB speeds (60-66MHz)
  • Pentium Pro in 1996; Frequencies (150-200MHz); FSB speeds (66MHz)
  • Pentium MMX in 1997; Frequencies (166-233Mhz); FSB speeds (66MHz)
  • Pentium II in 1997; Frequencies (233-450MHz); FSB speeds (66 and 100MHz)
  • Pentium III in 1998; Frequencies (450MHz-1GHz); FSB speeds (100 and 133MHz)
  • Pentium 4 in 2000; Frequencies (1.3-1.5GHz); FSB speeds (400MHz)

DRAM Types

  • FPM DRAM used with the first Pentium Processors
  • EDO DRAM was introduced in 1995 (used with majority of Pentium Processors)
  • SDRAM was introduced in 1996; here is the max theoretical bandwidth of each application (PC66=528MB/sec, PC100=800MB/sec, PC133=1.064GB/sec, PC200=1.6GB/sec)
  • RDRAM was introduced in 1999; here is the max theoretical bandwidth of each application (PC600=1.2GB/sec, PC800=1.6GB/sec, PC1066=2.132GB/sec - NOT AVAILABLE YET); but dual channel, used in majority of RDRAM-based motherboards, doubles the bandwidth!
  • ERDRAM will be the next standard sometime in 2002

So if we analyze this information, we come to one conclusion: Rambus certainly has NOT stifled innovation; in fact, just the opposite. Rambus has ignited a flame under the semiconductor industry! Since Rambus wholly entered the PC business in 1999, consumers have seen a significant increase in raw power!

  • Increase in processor frequency from 1993-1998 = 1566%
  • Increase in FSB speed from 1993-1998 = 122%

Since Rambus entered the equation:

  • Increase in processor frequency from 1998-2000 = 233%
  • Increase in FSB speed from 1998-2000 = 300%

In the past 8 years, the ONLY major change in either Front Side Bus or RAM has been this year, with the introduction of Rambus! From 1993-1999, both Intel and Memory Makers have kept the status quo and only marginally improved either of these two critical components...Rambus steps in and steps up the game!

Intel and AMD: Gorilla's Revenge

Now, I want to mention AMD and their role in this saga. Since Intel signed an exclusive contract with Rambus in 1996, AMD has been an outsider. AMD decided to license Digital's EV6 bus design, which at the time, was substantially faster than Intel's own design. AMD's implementation ran at 200MHz and continues to run at that speed today. This was twice as fast as Intel 100MHz design, but was trumped by the Intel/Rambus 400MHz design this year. AMD was at a loss and knew it. They had three options in front of them.

They could either follow Intel's lead, or try to design a new Rambus-enhanced system bus. This choice was far too costly at the time.

Option number two was to implement RDRAM, albeit with their 200MHz design which would effectively only enable single channel RDRAM, with room to spare. They could have gotten a new license from Compaq (formerly Digital) to upgrade the bus to 333MHz - which is the max this design can handle, I think. Or they could have implemented the dual channel Rambus architecture on their 200MHz design and get comparable performance as the Intel i840.

Finally, they could spearhead DDR SDRAM adoption and take "the road less traveled." DDR fits into AMD's current framework nicely since it has an effectively 200MHz design, matching the AMD system bus. Unfortunately, this third route was a lot more difficult then anyone expected. DDR is a Rambus technology, but without Rambus' blessing. This is because DDR is only a subset of Rambus' RDRAM spec, and until recently, wasn't supported by Rambus. So AMD had a hell of a time and is still struggling to properly implement DDR into their designs. When finally implemented, it will still suffer a beat down by Intel's dual-channel RDRAM solution!

But that is the saga of Gorilla versus Monkey. There's no doubt that in the future, AMD will implement a dual-channel RDRAM solution, but it will probably at the time when Intel is introducing a quad-channel RDRAM solution. Unfortunately, for consumers, AMD will always be a step behind Intel. Rambus should provide an equal playing field, but Intel's already extensive knowledge of Rambus technology should keep them ahead.

The Memory Makers, Rambus, and Intel

The memory makers have reason to hate Rambus. Rambus has effectively cut them out of the equation. Here's why. For the past 10 years, memory makers have been designing and producing DRAM their way. That is, they all meet in this organization called JEDEC where they set the standards. Those standards are based on what the DRAM makers want and let me explain this. The memory makers are like any other commodity businessmen, they produce commodities at the cheapest price available. There is no advantage of one design over another; they are all the same JEDEC spec'ed part, just physically made by a few manufacturers. Their best interests are keeping the status quo so they don't have to make (PP&E) property, plant, and equipment investments. They don't care if the industry needs a faster memory; their sole interest is making the JEDEC part THE CHEAPEST! Rambus doesn't care either, but they have a new revolutionary design that scales! The problem is, JEDEC is an Antitrust MONOPOLY, not Rambus. So, in the last 10 years, as the Memory Makers were going about business as usual, a small company in Mountain View, California was designing a memory system that could scale with the processors they were interfacing with. Intel realized that this little company could bridge the gap between processor and memory bottleneck that was created early on. Unfortunately for the Memory Makers, they were effectively shut out of the equation. Rambus stripped their whole bargaining power from them. Now, they just basically provide foundry services for Rambus. This leaves Rambus to design and develop all future next generation memory devices, based on what Intel tells them they need. Intel gets back into the memory game, albeit on its own terms. Rambus gets everything they want. And the poor memory makers are relegated to manufacturers. But that's what happens to companies who don't innovate! Their last ditch effort is to align with AMD, but that won't work.

The Horowitz/Farmwald Law

Moore's Law states that transistor counts on a CPU will double every 18 months. This scenario certainly hasn't been happening on the memory side of things. BUT, Rambus has a technology that can make this happen with memory as well. We can call this the Horowitz/Farmwald Law. And it states that as CPUs get faster, memory will keep pace. So if transistors double every 18 months, the interface between CPU and RAM will double every 18 months! It's a simple adaptation of Moore's Law for RAM.

We will finally have a harmony between the two most crucial parts of an electronic device, CPU and memory. So to get back to the point I was trying to make at the beginning, Rambus has NOT stifled innovation or hurt the consumer. Quite the contrary, Rambus has heralded in a new generation of memory interface that will enable system integrators and consumers alike to continue to enjoy, and in the case of RAM, bring us up to speed, on the multimedia world we live in!

This leaves only one question in my mind, what's the next problem Rambus will tackle?

Best Regards,

the-breadman