|
|
AMD is one of the longest-lived manufacturers of x86 computing technology. The company's market share increased rapidly after the release of its K5 core architecture. Some iterations of AMD designs were almost indistinguishable from the products of Intel or Cyrix. In the early days, nearly all x86 products resembled each other closely, as manufacturers worked to establish a common PC standard. The success of AMD's current Athlon platform shows that the company has learned to harness the power of market diversification that began with the original K5.
With the K5 series, AMD first diverged from its competitors in design, engineering, and production. The K5 contained one of the most efficient integer execution units ever made, though its lack of a full-pipelined floating-point unit essentially relegated K5 architecture to the budget realm. Gaming and multimedia interests drove the high-end market toward performance platforms, like Intel's redesigned Pentium processor, and away from architectures like the K5, which was closely related to the 486 design.
The AMD K5 compared favorably to the Pentium running a wide range of desktop applications. While the underpowered floating-point unit raised questions, performance was still acceptable for most users of the day. The real stumbling block for the K5 was not its engineering as much as the way it was sold to consumers.
AMD used integer benchmarking results compared to the Pentium to name, rate, and market each model in the K5 series. The choice to compare benchmark numbers instead of clock frequencies caused some analysts to accuse AMD of deception. While the K5 could outperform the Pentium "clock for clock" in integer applications, multimedia programs lagged. Many vendors decided to market the K5 under its operating frequency, while retaining the performance rating (PR) number for general comparisons.
The K6 was an evolutionary step for AMD. It offered minimal core improvements over the K5. The main architectural differences included MMX (multimedia) instruction set enhancements and adaptations to the core that supported a split-voltage interface to maintain motherboard compatibility with Intel's Pentium MMX. AMD's marketing strategy for the K6 was, however, dramatically different than it had been for the K5.
AMD chose to develop, market, and name the K6 processors by their operating frequencies, dropping the questionable PR-naming convention. Further core refinements, combined with additional MMX improvements, brought the K6 series a loyal following in the budget market segments. Intel retained the lead in multimedia performance. The rather dismal floating-point unit persisted in the K6 processors, limiting AMD to the entry-level realm.
Eventually, AMD recognized the need for an architecture revision and a new multimedia solution, especially after Intel released the powerful Pentium II platform. While the K6 could compete with the Pentium MMX, it could not compete with the P2 core architecture.
Lacking the resources to develop a completely new platform, AMD looked toward a new multimedia instruction set to compete with Intel's superior floating-point execution. The company adopted a streaming multimedia instruction set for its new K6-2. The K6-2 introduced key upgrades to the K6 design. The 3DNow! instruction set extended the floating-point capabilities of the K6, and an increase to 100 MHz in front-side bus speed improved memory and subsystem performance. The core die size also decreased to .25 micron, allowing the K6-2 to scale toward 550 MHz.
3DNow! was a forward-looking technology developed along the same lines as Intel's MMX instruction set. MMX allows for more efficient processing of integer mathematical computations; 3DNow! allows efficient combination of multiple floating-point operations in specialized instructions. 3DNow! also required many software optimizations at the programming code level, so AMD moved quickly to acquire and maintain development contracts.
When running properly optimized software applications using 3DNow! technology, instead of pure floating-point mathematics, the K6-2 could compete and even surpass the Intel architecture's performance. Game and hardware driver developers quickly adopted the new standard, and the K6-2 infiltrated all desktop markets. Even without 3DNow!, the K6-2 offered adequate floating-point performance through sheer increases in MHz (compared to the K5). Given low retail prices and compatibility with existing Socket 7 motherboards, the K6-2 rapidly emerged as the first real competitor to Intel's Pentium II and Celeron product lines.
Once the K6-2 found acceptance in the desktop market, AMD saw the need to develop a workstation and server-grade processor. The primary limitation of the K6-2 is memory bandwidth performance, but AMD could do little about motherboard chipset design. Manufacturers like VIA and SIS had failed to improve memory bandwidth or latency since the inception of Socket 7. ALI did improve the architecture related to memory access, but its chipsets suffered compatibility and stability problems when used with the newly popular AGP graphics accelerators.
AMD recognized the deficiency in bandwidth and opted for a solution at the processor level. The K6-2 implements a one-tier internal cache memory buffer, with the second-level buffer surface mounted on the motherboard. The Level 2 cache is accessed at the front-side bus rate because the chipset handles the transactions.
AMD's K6-3 took a page from Intel's playbook by introducing a Level 2 cache embedded inside the processor core. This internal L2 operates at the same rate as the core, increasing the cache memory latency and bandwidth exponentially compared to the K6-2. The cache memory mounted on the motherboard became a third-tier cache acting as an intermediary between the fast processor and slow system memory.
The K6-3 was marketed primarily as a server. Its new three-level cache architecture encouraged integer performance exceeding that of Intel's Pentium II at similar core clock rates. Curiously, the K6-3 450 could often exceed the integer performance of the more expensive Pentium III 450. Server environments have little need for floating-point multimedia performance; thus the K6-3 captured a significant slice of the higher-end processor market due to its superb performance and low-cost implementation. AMD could finally compete with Intel at all levels of the x86 market.
The plus series (marked with the + symbol) of K6 processors was a stopgap to maintain market visibility for AMD in the mobile computing segment. Both the K6-2 and K6-3 product lines featured a plus series variation. Core die size was reduced to .18 micron, which improved thermal efficiency. Architectural improvements include AMD's new 3DNow! DSP standards. The K6-3+ core remained basically the same, while the K6-2+ was upgraded to include an on-die 128 KB Level 2 cache memory architecture.
|
|