Pentium Pro

The Pentium Pro was a family of Intel x86 microprocessors introduced on November 1, 1995. It represented Intel's transition from the P5 architecture used by the original Pentium to the P6 microarchitecture, a design that became the technical foundation for a large portion of Intel's processor portfolio for more than a decade. Although the product name suggested a direct successor to the Pentium, the Pentium Pro was ultimately positioned primarily as a high-end processor for servers, workstations, multiprocessor systems, and technically demanding desktops. The processor's central innovation was its ability to translate conventional IA-32 instructions into internally buffered micro-operations. Those micro-operations could be analyzed, reordered, renamed, and executed speculatively, allowing the chip to perform out-of-order execution and exploit instruction-level parallelism. This was a major architectural departure from the more conventional in-order behavior associated with earlier x86 processors. The design used a 14-stage superpipelined structure, an instruction pool, register renaming, and multiple execution units. It also introduced conditional-move instructions to Intel's x86 instruction set and supported upgradeable microcode under BIOS or operating-system control. The Pentium Pro was manufactured in clock speeds of 150, 166, 180, and 200 MHz, with 60 or 66 MHz external bus speeds. A 133 MHz prototype existed but was not commercially released. The processor supported dual- and quad-processor symmetric multiprocessing and used Socket 8, a comparatively large package. Its most distinctive physical feature was an on-package level-two cache. Rather than placing the cache on the motherboard or integrating it directly into the processor die, Intel combined the processor die and one or more cache dies inside a ceramic multichip module. The cache operated at the processor's full core speed through a dedicated back-side bus, improving memory performance and reducing contention on the main system bus. Configurations were offered with 256 KB, 512 KB, and 1 MB of L2 cache. The product had important limitations. Its out-of-order design could perform poorly with the large amount of 16-bit and mixed 16/32-bit code common in contemporary consumer software, including portions of Windows 3.x and Windows 95. It also lacked the MMX instructions introduced in later Pentium products, which reduced its appeal for multimedia applications. The package's complex manufacturing process produced relatively low yields and contributed to high system prices. As a result, the Pentium Pro did not replace the original Pentium across the entire personal-computer market as initially envisioned. Its strengths were more apparent in 32-bit operating systems such as Windows NT, Linux, Unix, and OS/2, as well as in server and workstation workloads. The processor was also used in large-scale scientific computing. The best-known example was ASCI Red, a massively parallel supercomputer that became the first system to exceed one teraFLOPS and later held the top position in the TOP500 ranking from 1997 through 2000. Each ASCI Red node used a pair of Pentium Pro processors. Intel later refined the P6 lineage in the Pentium II and Pentium III families. The Pentium II Xeon became the principal successor for high-end server and workstation applications in 1998. Although the Pentium Pro itself was short-lived as a commercial product, its architectural concepts—including speculative out-of-order execution, register renaming, and the broader P6 design philosophy—continued through later Intel processors and influenced the Pentium III, Pentium M, and eventually the Core family.

Intel developed the Pentium Pro as the first production processor based on the P6 microarchitecture, replacing the P5 architecture used by the original Pentium. Fred Pollack served as the lead architect. The design was intended initially to cover a broad range of personal-computer applications, but its commercial role narrowed as Intel and the market recognized that its strengths were concentrated in 32-bit professional workloads rather than the mixed 16-bit software environment still common in the mid-1990s. The Pentium Pro was introduced on November 1, 1995, at clock speeds ranging from 150 to 200 MHz. A 133 MHz engineering version was produced but never released. Unlike earlier Pentium processors, the P6 core translated IA-32 instructions into smaller internal micro-operations. These operations were placed into an instruction pool and could be reordered and issued speculatively, enabling out-of-order execution and register renaming. The processor had a 14-stage pipeline, three instruction decoders, and several execution units for integer, floating-point, load, and store operations. Its 36-bit address capability, used with Physical Address Extension, allowed systems to address substantially more physical memory than conventional 32-bit addressing alone. Intel also introduced conditional-move instructions with the Pentium Pro. These instructions allowed certain conditional operations to be performed without a conventional branch, potentially improving performance in suitable code. The processor supported microcode updates controlled through firmware or operating-system mechanisms, another capability that became significant in later Intel designs. The cache system was unusual and technically important. The processor and its L2 cache were separate silicon dies assembled in one package. Initial versions included 256 KB of cache, while later versions offered 512 KB or 1 MB. The cache communicated with the processor over a dedicated full-speed back-side bus rather than sharing the main front-side bus. This arrangement reduced memory bottlenecks and was particularly valuable in multiprocessor configurations. It also made manufacturing difficult: defects in any die could cause the complete multichip package to be rejected, contributing to cost and yield challenges. Pentium Pro systems were commonly configured with two or four processors. The first computer system reported as shipping with the processor was Advanced Logic Research's Revolution Quad6, released in April 1996. Intel did not produce a conventional mobile version because of power and thermal constraints, although at least one vendor marketed a portable system using the chip. The processor's performance profile was mixed. On suitable 32-bit integer workloads it was highly competitive and could outperform contemporary processors with lower clock-for-clock efficiency. However, frequent use of partial registers and legacy 16-bit code could cause pipeline disruption. This mattered because much contemporary desktop software retained 16-bit components, even when operating under a nominally 32-bit operating system. The Pentium Pro also lacked MMX support, limiting its performance in emerging multimedia applications. Slow video-memory writes were another weakness in early graphics workloads, although software and driver techniques later improved performance. The processor achieved a stronger legacy in servers, workstations, and supercomputing than in ordinary consumer PCs. ASCI Red, built with Pentium Pro processors in dual-processor nodes, exceeded one teraFLOPS and held the top position in the TOP500 supercomputer list from 1997 to 2000. Intel subsequently refined the P6 approach in the Pentium II and Pentium III generations. The Pentium II Xeon became the direct high-end successor in 1998. P6-derived concepts continued in later mobile and mainstream designs, including Pentium M and Core-related processors, making the Pentium Pro an important architectural ancestor even though the branded product had a relatively short market life.

• 1998

  Pentium II Xeon becomes successor

  [1]Pentium Proen.wikipedia.org

  Intel's Pentium II Xeon succeeded the Pentium Pro in high-end server and workstation roles.

• 1997

  1 MB cache configurations appear

  [1]Pentium Proen.wikipedia.org

  Intel expanded the Pentium Pro cache range to include configurations with up to 1 MB of on-package L2 cache.

• 1997

  ASCI Red reaches the top of TOP500

  [1]Pentium Proen.wikipedia.org

  ASCI Red took the number-one position in the TOP500 ranking, retaining it through 2000.

• 1996

  Quad-processor systems enter the market

  [1]Pentium Proen.wikipedia.org

  Advanced Logic Research released the Revolution Quad6, one of the first systems designed to support up to four Pentium Pro processors.

• 1996

  ASCI Red deployment

  [1]Pentium Proen.wikipedia.org

  ASCI Red used paired Pentium Pro processors in each computing node and became the first computer to exceed one teraFLOPS.

• 1995

  Pentium Pro introduced

  [1]Pentium Proen.wikipedia.org

  Intel introduced the Pentium Pro on November 1 as the first commercial processor using the P6 microarchitecture.

High-end server, workstation, scientific-computing, and professional desktop processor

• Microprocessor · 1995

  Pentium Pro 150

  The 150 MHz entry version of the commercial Pentium Pro range. It used the P6 architecture, Socket 8 packaging, and an on-package L2 cache. Like other Pentium Pro models, it was intended for professional desktops, workstations, servers, and multiprocessor configurations rather than cost-sensitive consumer systems.

• Microprocessor · 1995

  Pentium Pro 166

  A 166 MHz Pentium Pro model produced on Intel's later 0.35 micrometer processor-die process. It offered the same P6 out-of-order execution model and Socket 8 platform as the other mainstream versions, with cache capacity varying by product configuration.

• Microprocessor · 1995

  Pentium Pro 180

  The 180 MHz version targeted high-end desktop, workstation, and server systems. It combined the P6 execution engine with a full-speed on-package cache and support for symmetric multiprocessing, making it suitable for operating systems and applications that could exploit multiple processors.

• Microprocessor · 1995

  Pentium Pro 200

  The fastest standard Pentium Pro model, clocked at 200 MHz. It was sold in configurations with different cache capacities and was frequently used in premium workstation, server, and multiprocessor systems. Its high cost and limited advantage with legacy code restricted its mass-market appeal.

• Cache configuration · 1997

  Pentium Pro 1 MB L2

  The largest Pentium Pro cache configuration used two 512 KB cache dies alongside the processor die in a multichip package. Its full-speed dedicated cache path improved performance in memory-intensive professional and multiprocessor workloads, but the additional dies increased manufacturing complexity and cost.

• 1995

  Narrow the Pentium Pro toward professional computingStrategy

  Background

  The product was initially conceived as a broad replacement for the original Pentium, but its difficulty with common 16-bit and mixed 16/32-bit software made it less attractive for mainstream personal computers.

  Action taken

  Intel positioned the Pentium Pro more strongly toward servers, workstations, high-end desktops, and other environments running fully 32-bit operating systems.

  Aftermath

  The narrower positioning aligned the product with its strongest technical characteristics, while the later Pentium II addressed some of its consumer-market limitations.

  Source

• 1995

  Adopt a separate on-package L2 cacheProduct launch

  Background

  Large integrated caches were difficult to manufacture economically at the time, while motherboard cache shared the main system bus and created a performance bottleneck.

  Action taken

  Intel placed the processor and cache on separate dies within a ceramic multichip module and connected them through a dedicated full-speed back-side bus.

  Aftermath

  The design delivered substantially faster cache access and strong multiprocessor performance, but its multi-die production process reduced yields and contributed to high prices.

  Source

• 1998

  Pentium II Xeon succeeds Pentium Pro in high-end systems[1]Pentium Proen.wikipedia.org

  Product generation

  Intel's Pentium II Xeon, derived from the P6 design lineage, became the successor for high-end server and workstation applications.

• 1996

  Pentium Pro becomes the basis of ASCI Red[1]Pentium Proen.wikipedia.org

  Other

  Pentium Pro processors were selected for the ASCI Red supercomputer, which became the first computer system to surpass one teraFLOPS and later led the TOP500 rankings.

• 1995

  Intel introduces the Pentium Pro[1]Pentium Proen.wikipedia.org

  Product launch · Product generation

  Intel launched the Pentium Pro as its first commercial processor based on the P6 microarchitecture, targeting demanding desktop, workstation, and server applications.

• Pentium Pro floating-point conversion erratum documented[1]Pentium Proen.wikipedia.org

  Other

  A floating-point unit issue, commonly called the Pentium Pro and Pentium II FPU bug or the flag erratum, was identified in a narrow floating-point-to-integer conversion case. The problem was considered unlikely to affect most software.

1. 1Pentium Proen.wikipedia.org