dacs.doc electric

THE FEBRUARY DACS general meeting introduced us to the dazzling new technology of the AMD Athlon chip. We had all heard about it but probably dismissed it as a knockoff of the real thing (that being Intel, of course). But walking out of that meeting were a lot of people with radically changed attitudes.

From AMD's office in Southbury came Steve McKey, District Sales Manager, and Mauro Lupero, Senior Field Application Engineer. Steve gave us a perspective on AMD as a company and an introduction to the Athlon chip. Most of us were surprised to hear that AMD has been around for 30 years and now has a significant share of what we generally assume to be Intel territory. The major point of Steve's introduction, though, was to assert that AMD's new Athlon chip is not a cheaper substitute for the Pentium III chip but a better-performing, in-your-face challenger. AMD must now be considered a serious player on Intel's level.

In taking on Intel, AMD has faced formidable obstacles. They have to build chips that not only do everything that the Intel chips do but add something extra. Like the situation where Ginger Rogers had to do everything that Fred Astaire did--and do it with high-heeled shoes, AMD is faced not only with taking on the glamour chip and making one more appealing, but with persuading entrenched Intel fans that their product performs just as well if not better and costs about the same.

With all the technoglamour of today's microprocessors, we tend to forget that basically they do the same thing that computers such as the IBM 704 did 35 years ago with vacuum tubes. Then and now the main objective is to get bit patterns from memory into the central processor's internal registers, do something with them, and then get the altered bit patterns out of the registers back into memory or to external devices. Basically this is one of the main objectives of a house: to get water into the house, do something with it, and then get it out of the house.

The drill works like this: Read a bit pattern into a special register, and then interpret it as an instruction to (a) read another bit pattern into a different register (data register); (b) do something with one or more data registers (arithmetic, logic, etc.); or (c) write the bit pattern in a data register back to memory or some peripheral device.

Even though the basic operation hasn't changed, chips sure have gotten faster over the past 35 years. Their increased speed has come from two factors: the electronics are speedier, and the central processor does more things at once (in parallel).

Mauro Lupero spent most of the meeting explaining how the Athlon chip is doing the drill better than Intel's Coppermine Pentium III chip. He showed us comparative diagrams of Athlon and Pentium III chip architectures, comparing the Athlon's 128KB L1 cache to the 32KB one in the Pentium.

The Athlon L2 cache can be up to 8MB, but it is off-chip. The Pentium L2, on the other hand, is 256KB but is on the processor die and inherently faster. Later versions of the Athlon will move its L2 on-chip, but it certainly won't be anything like 8MB.

The fastest Pentium system bus is 133MHz, but the Athlon's EV6 technology bests it at 200 MHz. AMD licenses the EV6 from Digital, which used it on its Alpha systems. The Alpha system, of course, is faster than anything, but it does things its own way. If Ginger had tried that, she would have been out of a job, the same penalty that the Alpha paid. AMD's incorporation of the EV6 bus technology, however, is behind the curtain, so while it helps to get the water in and out of the house faster, it doesn't violate "compatibility."

To make the computer "do something with the registers" faster, computer architects over the past 35 years have added circuitry to accomplish as many things in parallel as possible. Multiple units for arithmetic and address processing are on-chip and supported by prediction units, schedulers, and control circuitry that determine data dependencies and even do speculative execution.

Mauro proudly described to us the Athlon's three pipelined floating-point units, three integer units, and three address calculation units. This is more than twice the parallel processing potential of the Pentium III chip.

To add glitter and steal the show the way Ginger could with her smile and her costumes, AMD has added additional instructions for high-performance graphics operations. Called 3DNow!, there are 21 instructions from AMD's K6 series and 24 additional instructions new with the Athlon. These instructions are not on the Intel chips. They manipulate 3D-graphics data structures. Programmers writing 3D apps can use these instructions in their source code for better effects than Intel chips can provide. This means the application developers must write both AMD-specific and Intel-specific code segments. At run time this is detected by the system, and the appropriate driver is loaded for the processor to use. That the major software houses are doing just this is the real mark that AMD has come of age and is in open competition with Intel.

And there are more AMD innovations that Mauro told us about. The company's operation in Dresden, Germany, will soon be producing Athlon chips using copper on silicon rather that the traditional aluminum. Developed by IBM, this major technology breakthrough promises even more blinding speed.

Will Ginger make it on her own? We certainly hope so. Two strong competitors are good for the consumer (us), good for the industry, good for the national economy, and even good for the competitors themselves. At our February meeting both Steve and Mauro made a convincing case for AMD's potential. Now we must do our part by following the literature and the Web pages, particularly Mauro's favorite www.tomshardware.com. Then, when decision time comes around, we must seriously consider the alternatives.