Sure, the Pentium II is fast, and the PowerPC G3 is even faster, but the lineup of chips coming soon will blow you away. The 64 bit Merced processor from Intel is expected to be released sometime late next year, and will represent the first time the company has strayed away from the original x86 architecture in years. Also expected to make its debut in the next few years is the Power PC G4, the successor to the G3, which has an even more promising architecture than the Merced. Even before this battery of new chips is released, however, the existing chips' speeds will be pushed to the limit; Intel has plans to take their Pentium II to 450 MHz by 1999, and Motorola/IBM will tweak the G3 to 500 MHz.
The Intel Merced processor achieves higher performance not by executing instructions faster, but by executing more instructions concurrently, a feat called parallelism. The technology that drives this concept is called EPIC (Explicitly Parallel Instruction Computing) named for its ability to maximize parallel execution. Working hand in hand with the new 64-bit instruction set, Intel has already created prototypes of this revolutionary 64-bit processor.
There are many advantages to the Merced processor over previous Intel chips. It is scalable, so additional execution units can be added to increase the number of instructions executing simultaneously. In explicit parallelism, the chip's compiler organizes the code efficiently, so the processor can be more effective handling the data. It will no longer have to redirect instructions and thus will be faster. Branches, which represent a decision between two sets of instructions, also hamper the potential processor speed in all microprocessors. While current architectures mispredict only 5 to 10 percent of the time, these errors can reduce speeds by as much as 40 percent. During idle time, the Merced chip executes both instructions of a branch, so by the time a misprediction does occur, the other instruction has already been started. This harvests more of the potential speed of a processor.
The technical aspects of this new processor are quite impressive on paper. Boasting a 64 bit architecture, parallel executions, and the advanced concepts of prediction and speculation, it seems Intel will be unmatched in raw processor power for years to come. Or will it? Scientists, executives, and other technology analysts may not be quite as optimistic as Intel employees. As Keith Diefendorff, an Apple engineer, stated: "I would cancel the whole Merced program. I would milk that [32-bit] architecture for another 25 years." He also said that the performance gained by creating a new chip architecture would be surpassed by performance gains from moving to smaller feature sizes - even when applying these new manufacturing techniques to chips using the current architecture. Bruce Lightner, Vice President of Development at Metaflow Technologies agreed. "Users haven't even fully adopted 32-bit architectures. We're still running 16-bit code." To run Intel x86 code, the Merced processor would require a virtual instruction set. While this will not slow the processor tremendously, the die area used by this instruction could have been imprinted with instruction sets for faster native code. Even if the processor is a success, a problem noted by Martin Reynolds of Dataquest is that DRAM speeds cannot keep up with the processor speeds. With these "super-fast" processor being introduced, the memory will hamper the speeds of the computer. This will effectively limit the speed of these new processors, making them only as fast as the memory they utilize.
This transition to 64 bit chips, if marketed well and made cheap enough could be a great success for Intel. In the event that it is not, however, this could be their undoing, giving competitors an edge. However, nobody should contest that this chip will operate at spectacular speeds as promised.
The 64-bit Alpha architecture from Digital Equipment Computing is also going through renovation. The DEC Alpha microprocessor is currently the fastest processor in the world, and Digital is attempting to retain the lead over Intel and IBM for the next few years by designing chips to compete with the other planned speed demons. Digital claims that the third generation Alpha 21264 family will deliver up to five times the fastest performance of any currently available architecture. This new processor will run UNIX, OpenVMS and Windows NT; and could mean trouble for Intel. The new Alpha will utilize 0.35-micron process technology when it first ships this summer. Each chip will contain 15.2 million transistors, twice as many as in the Pentium II. The Alpha 21264 will use intuitive execution for improved branch prediction - a concept similar to the one which will be used in Merced. High-speed access to level 2 cache and system memory will also be integrated into Digital's Alpha motherboards.
While many of the same technologies listed above will be used in Intel's Merced, the Alpha 21264 will be shipping nearly a year before the Merced's introduction. The one large drawback to the Alpha is that it will not be a cheap home-consumer chip. It will be installed into high end servers and workstations - a relatively small market. Other than that, prospects are good for this 64-bit wonder. It will ship soon, and has a multi-year history of being the best.
The IBM/Motorola PowerPC G4 will also implement a revolutionary design. Instead of designing an entirely new architecture, this corporate duo decided to take multiple chips similar to the G3 and link them together inside of one microprocessor. The G4 will also use the IBM copper process innovation, which will add to the speed. This essentially tricks the operating system into thinking that the G4 is one microprocessor, while really it consists of many slightly modified G3 chips. The possibilities of this design are unlimited. Intel did not have this same option as IBM and Motorola though, for the Pentium II chip draws eight times the power of the G3, runs much hotter, and is less efficient. While this chip has a lot of potential, only Mac OS and selected UNIX systems will be able to use this chip - while Intel caters to the mainstream PC market.
Although word is scarce about the technical aspects of this processor, an industry rumor site said indicated that Motorola has developed a prototype G4 PowerPC processor running at 840Mhz. The source reported that the creator of the system estimated its speed at "probably three times faster than any G3 box will ever be." Other sources said that megahertz-for-megahertz, a G4 should be two or three times faster than the G3. The processors will be released in 1999, and will run at speeds between 650Mhz and 1.2Ghz. Rumor has it that it will be based on a 64-bit processor core and will support larger cache capabilities than the G3.
Which chip will be the fastest? Which will sell best? The answer to the former is probably the G4, which will probably have unlimited speed potential - at a cost. Rumors that the chip will be nearly twice as expensive as the Merced chip have been circulating. If IBM and Motorola can overcome the expensive production costs in some way, this chip will have a much brighter future. The Merced chip will most definitely sell - with Intel's history of excellent marketing, and with the enormous market share of Microsoft Windows PCs, Intel can't go wrong. It is too early to tell, however, whether or not 64-bit computing will catch on because of the Merced chip introduction. In the past, Digital Computing's Alpha chip has proven to be too expensive for a more mainstream market, although that may soon change with the introduction of the new chip this summer. All of these chips have one thing in common; they are pushing the speed limit, and will make today's machines obsolete. One must remember when looking to purchase a computer, though, that the first machines using these new chips will be rather expensive - and for common applications, the computers of today run just fine....