Choosing A Chip

This Number Soup Used To Be A Lot Simpler

Choosing a chip for a new computer system used to be a much simpler task. You decided if you were an "Intel Fanatic," or an "AMD Fanboy" and you bought the fastest clock speed that your budget allowed. Before you could say "64 bit processing," you had a great new chip being FedExed to your front door.

Along the way, the chip manufacturers decided that clock speed was not he only way to put more ponies under the hood. At least the systems ran cooler by doubling the cache rather than adding another gigahertz to the clock speed. However, this created the dilemma of how fast the chip really was. After a failed start by Cyrix (because of inflated ratings), AMD was able to get their PR rating system to stick. When you bought an Athlon XP 2200+, you were getting a chip that ran close to the Pentium 4 2.2 GHz part. If anything, AMD underrated their chips which fostered acceptance as nobody felt like they were getting hoodwinked.

Up until this point, things were still clear, and easily comparable. Intel then decided to revamp their nomenclature into a series of numbers. Interrelated chip technologies, such as 64 bit support, front size bus speeds, and most recently, dual core chips have also muddied the performance waters, and complicated performance comparisons.

This has created some fascinating performance rating issues. For example, AMD has produced three 3000+ chips: the Athlon XP, the Athlon 64, and the Sempron. By their own PR system, they are implying that they all run equivalent to a Pentium 4 3 GHz part. On the other hand, I doubt that anyone would think that a Sempron 3000+ can run neck and neck in most tasks with its distant relative, the Athlon 64. Another issue is that AMD has been assigning PR equivalents to parts that simply don't exist in Intel's stable. AMD's 4800+ X2 dual core chip is the perfect example of this. Intel has never even released a chip that runs at 4 GHz or faster. Outside of an overclocker's dream system, there is no 4.8 GHz chip from Intel. How the heck is AMD going to know that their chip can keep up with it? Finally, AMD adds 200 to the PR of the chip when they add cache, or increase the clock speed. In benchmarking, the clock increase usually increases the performance more than the extra cache does. Clearly, this is not the smooth step wise performance increases between chips that they want you to believe.

Intel adopted their own number system to downplay clock speed, and account for other performance variables such as cache, front side bus, Hyperthreading, and dual cores. The chips are then categorized into number based categories of 3xx, 5xx, 6xx, 7xx, 8xx, or 9xx. The idea is that the higher the number category, the faster chip. Also, within each category, the faster the chip, the higher the overall number. There is even more potential for confusion here as the numbers are quite arbitrary. Also, having a "D" designates a dual core in the eight series, but not in the three series. Furthermore, the eight series should be a lot quicker than the series below it. However, I doubt that the Pentium 805 D will be any faster than a Celeron 355 D on single application tasks that don't take advantage of the 805's dual cores.

So where does that leave us? When buying a chip, we need to research all the variables mentioned earlier to really know what we are getting. Finding benchmarking results for the chip is also key as actual performance is far better than theoretics. And if you can overclock your chip to 4.8 GHz, that should produce the desired result!