Opteron Flies, but AMD Hits Speed Bumps

Taschek: AMD had a golden opportunity to succeed here, but its poor execution over the last year has hampered its chances for success.

With Opteron, the best we can hope for is that the marvelous 64-bit processor wont be buried by AMDs marketing bungles. Unfortunately, it looks like AMD is still bungling, much to the chagrin of the IT community.

AMDs rationale for 64-bit computing is in line with most industry thinking. Sixty-four-bit systems are necessary because the volume of data that systems must address is doubling each year while memory prices are decreasing by half each year.

To succeed with Opteron, AMD had to execute three tasks. The first was to cure the Intel-inspired megahertz madness and still produce faster processors than Intel. Just a few years ago, much of the desktop-processor-buying public upgraded systems on a megahertz basis. If 500MHz was too slow, consumers thought getting a 1GHz system would double performance.

As youd expect, AMD says the 64-bit Opteron will be faster than the 64-bit Itanium.Perhaps more surprising is that AMD also claims that the 64-bit Opteron is faster at 32-bit operations than Itanium and the fastest Xeons, including those that use the 533MHz front-side bus (the bus that connects the processor to the memory controller and along which all memory operations need to travel) and perhaps even the 667MHz bus design expected next year.

In servers, its not about the megahertz, as Intel realized a couple of years ago, when it released the PIII Xeon, basically a Pentium II core with a few megabytes of cache integrated on the processor die. This played into the hands of AMD. Both Intel and AMD turned to architecture advancements as their main differentiators. Intel came up with Hyper-Threading Technology—a brand name for Intels ability to execute multiple application threads in parallel. At that point, Intel shifted the megahertz battle from the processor to the processor bus.

AMDs answer to hyperthreading technology is called HyperTransport—AMDs implementation of an open specification governed by the HyperTransport Consortium. Instead of jumping on the bus battle with Intel, AMD claims HyperTransport obviates the front-side bus and gets rid of latencies in SMP systems. AMD is basically telling Intel that its not getting into the bus race because buses are a poor form of transportation in the age of flight. Result: AMD succeeds.