40 Years of 80386: Intel's Most Important Product

Contents

40 years ago, Intel introduced its first true 32-bit processor, which back then didn't yet have a fancy marketing name. Simply dubbed "80386", this CPU was built in numerous variants with extended or limited functionality until 2007 – and not just by Intel. However, the beginnings of the architecture known today as "x86" were bumpy.

In 1985, Intel was under considerable pressure in the PC boom, which had particularly impacted professional environments. IBM had already specified for the first PC, the model 5150 from 1981, that processors had to come from multiple manufacturers. The company considered a single supplier as "single source" to be too risky. Thus, Intel granted numerous licenses for replicas, which was not to be the end of it. IBM itself manufactured the 8086 and its successors 8088 and 80286, as did other companies like Fujitsu, Mitsubishi, NEC, Siemens – and of course AMD.

Intel almost misses the 32-bit connection

These companies did not limit themselves to copies of the 80x86 CPUs but expanded the designs and increased clock frequencies, so Intel urgently needed an entirely new architecture. This architecture was intended to work with 32-bit addresses and data, as with Motorola's 68000 series CPUs. Those 68000s enabled a new class of home computers from 1984 with the Apple Macintosh and in 1985 with the Atari ST and Commodore Amiga, which were much cheaper than PCs and yet powerful. Although the arithmetic logic units (ALUs) of the 68000 were still 16-bit, the 68020 already compensated for this minor drawback in 1984.

Intel's initial hope was a design called iAPX 432, which seems somewhat crude by today's standards. It was 32-bit but split across two chips, far too complex, and with a completely new instruction set. However, the main selling point for the PC had always been backward compatibility. Furthermore, the 432 couldn't get going in terms of performance; in parallel, Intel developed the 80386 as a 32-bit extension.

The Placeholder

The 80386 was originally intended as a placeholder to bridge the time until the hoped-for success of the 432, as one of the two 386 designers recalls – a certain Pat Gelsinger, who decades later would become CEO of Intel, and last year was put into the company's now greatest crisis. He describes the 386 development in detail in his guest article "For ever young", (c't 13/2003, pp. 90-95).

Besides Gelsinger, John Crawford was responsible for the 386. The standard work "Programming the 80386" also comes from the two of them, which was only published in 1987 – practical when chip designers also take care of the software. A year earlier, Intel itself had already released the "Programmer's Reference Manual" (PDF), but it is more of a documentation of all functions and less of a practical manual.

Based on the 80286, Crawford and Gelsinger not only brought 32-bit wide registers, address, and data types to Intel's architecture, which were also routed externally as buses. So that the then unimaginably large address space of 4 GByte could be used meaningfully, there were several addressing modes as well as an instruction set expanded by a whole 54 instructions. The addressing modes, along with hardware memory management (MMU), proved to be the cleverest, enabling larger operating systems and applications. And a three-stage pipeline speeds up instruction execution.

Protected, Flat, or Real – Plenty of RAM for Every Purpose

With the 80286, Intel had already introduced the so-called Protected Mode, which could isolate memory areas from each other for applications. However, it was quite cumbersome to program. For the 80386, Real Mode for DOS compatibility with multiple instances and "Flat Mode" for large memory models were added. This already allowed a bit of multitasking at the hardware level. However, the processor's capabilities were only gradually explored through memory managers overlaid by DOS.

The capabilities of the new architecture were so enticing that the first PC with an 80386 was not from PC inventor IBM, but from competitor Compaq. Their Deskpro 386 model also bore the name of the new superchip in its product designation; previously, PCs beyond the 8086 had been marketed as "XT" (Extended Technology, 8088) or "AT" (Advanced Technology, 80286) with these IBM designations. The deal with Compaq brought Intel's own processor into the spotlight within the entire environment of Intel clones and further developments. A "386" subsequently became synonymous with an entire PC category. The 20 MHz version of Compaq's supercomputer cost a hefty 16,000 Marks in 1988, even in its basic configuration with 1 MByte RAM.

The Fastest PC Doesn't Come from IBM

The first Deskpro 386 appeared in September 1986, almost a year after the 80386 was introduced. Initially, it struggled with bugs and low yields in chip manufacturing, so instead of the hoped-for 16 MHz, only around 12 MHz were achievable with some batches. Some models were even only approved for 16-bit software because the processor made mistakes when multiplying in 32-bit. Unlike the infamous FDIV bug, Intel documented this with inscriptions on the processors themselves. These problems were overcome just under a year later, but led to Intel's first red numbers in its history for the 1986 financial year. The 80386 therefore absolutely had to be a success.

In 1986, Compaq CEO Rod Canion told the magazine Computerworld that there was no point in waiting for IBM. And the company didn't want to wait any longer for an MS-DOS that fully supported the 386. So the Deskpro was launched with Microsoft's Unix derivative Xenix, initially primarily as a server. However, those who developed software themselves quickly realized what was possible with the 386 PCs. When IBM finally got around to it in 1987 and launched its 386 as the PS/2 – mind you, without the processor number in the name – many other companies had long since occupied the market. Much of what IBM devised for the PS/2, such as the Microchannel expansion cards, could not hold its ground for long.

Windows/386 with DOS Multitasking

At the end of 1987, Windows 2.0 was released, which was also available as Windows/386 with the corresponding DOS. This supported the 80386's Protected Mode to the extent that multiple DOS instances could be launched from within Windows. While not particularly stable, even later with Windows 3.1 and add-ons like Sidekick, it was a game-changer for some applications, such as programming or media creation, to run multiple DOS programs simultaneously. Numerous other applications also appeared as 386 versions, but interested parties had to wait for IBM for those as well. It wasn't until 1992 that their OS/2 2.0, which fully supported the 80386, was released.

In the years following the market launch of the first fully functional 80386 at 16 MHz, Intel increased the clock speed to up to 33 MHz, a doubling – unthinkable today. Other manufacturers like AMD and Chips & Technologies reached up to 40 MHz through extensive reverse engineering. At the end of the 386 era, and before the introduction of the 80486, the competition achieved a larger market share than Intel itself.

A 386 for Every Application

However, the then-leading processor did not remain idle during the huge boom of 386 PCs. The first 80386 was renamed i386DX because the cost-saving version i386SX was released in 1988, which only had a 24-bit address bus and could thus address 16 MByte of RAM. The data bus is also only 16 bits wide, while the instruction processing and registers are 32-bit, making them fully compatible with 386 software. And in 1990, the i386SL with partly static design was released. It was quite power-efficient and could put parts or the entire processor to sleep, which the new "System Management Mode" (SMM) allowed – again, this required operating system adjustments. And the 80387 coprocessor conquered the technical world. This Floating Point Unit (FPU) accelerated floating-point operations by several factors if the software was compatible – among other things, CAD became truly feasible on desktop PCs this way.

Overall, in just five years after a slow start, the PC industry had created an ecosystem in which 386 processors played a central role. From servers to mobile computers. Eventually, it was even so mobile that an embedded version (80386EX) was found in the Nokia Communicator 9000, that large flip phone with a keyboard, fax function, and email access – a sensation at Cebit 1996, mind you: eleven years before the first iPhone.

From IA-32 to x86

In the late 1980s and early 1990s, it became clear that PC processors could be well integrated into various application areas. And that this didn't require a single industry giant like IBM – even Intel was far from its later, and now lost, dominance. Nevertheless, the 80386 was so important to Intel that the company later renamed its architecture "IA-32" (Intel Architecture, 32 Bit), or x86 for short. Even modern x86 processors, even with 64-bit extensions, which first came from AMD, are still based on IA-32 at their core. This makes the 386 Intel's most important product of all time. Today, various ARM variants are increasingly competing with x86, from smartphones to notebooks to AI data centers.

<ir_inline itemname=inline_branding type=0> (nie)