Merced: Intel's 64 bit revolution

As with any other significant advance in PC technology, the planned launch of Intel's next generation processor, Merced, has caused huge amounts of confusion among users. In this particular case rumour, speculation and overzealous reporting have muddied the waters to an almost unprecedented level of misunderstanding.In a series of articles in this week ZDNet will seek to clarify the Intel roadmap for Merced.

Merced in a nutshell

So what exactly is Merced? At the moment, all of the Intel processors we use (PII, PIII, Celeron, Xeon etc..) are based on Intel's 32-bit processor architecture, that is IA-32. Merced will be the first processor to use the new 64 bit architecture -- IA-64. Without going into too much technical detail, this means that the data set, instruction set and address bus will all be 64 bits wide. That means the processor can address much more memory and process many more instructions far quicker than a 32 bit system.

The essential point to grasp here is that unlike Pentium III, which is essentially a Pentium II with a handful of extra features bolted on and a faster clock frequency, Merced will be a completely new processor design. For this reason, existing chip packaging and connectors, such as Slot 1, or Socket 370 will not be adequate for Merced and the chip will use a new system called Socket M.

Hey, that's EPIC man...

A major new technology featured in the Merced processor is called EPIC (Explicitly Parallel Instruction Computing), something which you're bound to hear a lot more about as we approach the chip's debut. A traditional processor wastes a lot of time trying to predict which instructions a program wants it to execute next. The processor is trying to save time by executing instructions in parallel, but it has to guess which instructions to execute because it doesn't know exactly what the program wants. The idea behind EPIC is that this "parallelism" is figured out when the program is initially compiled, cutting out wasted time on guesswork. In effect it gives the processor more explicit instructions on how to execute instructions in parallel, hence the term "explicit parallelism".

Another problem which has traditionally caused performance degradation is memory latency. Processors run much faster than memory hardware, and often the processor is left hanging around waiting for data to arrive from memory before the system can continue running a program. IA-64 will use a technology that Intel calls "speculation". Speculation orders memory to deliver data in advance, rather than waiting until it needs the data before asking for it. By doing this, data should arrive just as the processor needs it, thus reducing delays.

I want performance

The point to bear in mind about such features is that while Merced offers full binary compatibility with IA-32 software (yes, you can run all your existing Pentium based software), to get any serious performance increases programs will have to be recompiled using compilers designed to take advantage of the processor's benefits. This is discussed in more detail in our Merced Software article. However, all this doesn't mean that when Intel finally releases Merced (in the second half of next year, most likely) we'll all be able to go out and buy hugely powerful 64 bit desktop PCs. Intel plans to develop and market its IA-32 processors for desktop PC's well into the future. Initially and for a considerable length of time after Merced's launch, IA-64 processors will only be found in high end server applications.

Rather than bringing a new level of performance to the desktop, as some people seem to think, Merced's purpose is essentially to give Intel a stronger foothold in the enterprise computing market. It will be a very long time before we begin to see IA-64 based systems filtering down to our desktops and longer still before it's used in any sort of consumer applications.