The 486 line
Officially known as the Intel 80486, the 486 was the first x86 chip to feature pipelining, integrated math coprocessor, L1 cache, and burst-mode memory transfers. The notable addition to the instruction set was CPUID, which allowed for identifying the processor without resorting to hacks to detect obscure processor behavior. Later models also introduced clock multiplication, in which the core of the CPU ran at a multiple of the motherboard bus (what would later become known as front-side bus) speed.
The original 486s were the DX models, introduced in April 1989 with a traditional 1x multiplier and running at 5V, with 1.2 million transistors. It had 8 KB of write-through L1 cache, and no integrated L2 cache. Motherboards which provided for L2 cache also controlled it themselves.
The 486SX and 487SX were introduced in April 1991. The 486SX chips had the same specs as the previous DX series, though with a reduced 1.185 million transistors, to account for the loss of the integrated FPU. Early 486SX models had a present but disabled FPU, causing speculation that Intel was cutting losses by selling their DX chips that had broken FPUs as SX. While the cripple-and-sell-anyway philosophy would eventually become an integral part of the CPU business, later SX chips were actually a separate design with no FPU at all. The 487SX is discussed further below.
The 486DX2 (actually printed as 486DX2) was next in the line, in March 1992. The only difference between it and the DX was the introduction of a 2x multiplier. The similarly unremarkable 486SX2 followed two years later, in April 1994.
Meanwhile, November 1992 marked the introduction of the 486SL, a low-power 3.3-volt CPU aimed at mobile computing. It didn't live long, as most of its features were absorbed back into "SL enhanced" versions of the other processors. The system management interrupt from the 386SL was used as a building block for system management mode, which allowed for selectively powering down peripherals that weren't in use. The 486SL also introduced the familiar suspend feature.
The 486DX4 came in February 1994, offering higher multipliers, a lower-power 3.3V design, and a larger 16KB L1 cache to offset the penalty of the increased speed. A basic 1x-multiplied 486 averaged 2 cycles per instruction. Later 486s kept this ratio internally, but as they were at a multiple of the bus speed, they placed more demand on memory in the 5-10% of cases where the cache missed. The DX4 allowed for multipliers of 2x, 2.5x, and 3x. The larger cache raised the DX4's transistor count to 1.6 million.
The 487 was not really a coprocessor at all. It was a 486DX with a different pinout, including an extra pin which allowed the motherboard to detect its presence. If there was a 487 installed, the "main" processor was completely switched off, and the 487 took over all processing, using its integrated FPU for math. As weird as this may seem, things get even stranger.
Intel later offered their own upgrades for 486SX machines: the DX2/OverDrive and the much later Pentium OverDrive chips. The Pentium parts were 5V chips with 3.1M transistors, 32 KB of write-back L1 cache (split half-and-half for instructions and data), and 2.5x multiplier. Although they were advertised as fitting the 487 socket, Intel also manufactured OverDrives to fit the normal processor socket.
The OverDrives were created partly to compete with the blossoming market of their competitors' offerings, but due to limited speeds (both core and bus), failed to deliver the performance they promised. In particular, the Pentium's dual pipelines and faster ALU left it behaving like an extremely highly multiplied 486, which the increased cache alone couldn't compensate for.
Motherboards of the 486 era generally ran at 25-50 MHz bus speeds, and offered 16-bit ISA slots and some VESA slots. High-end systems in the later days had the much saner PCI bus. As noted above, some boards also offered external L2 cache. Memory was provided by either 30-pin (4 per bank) or 72-pin (1 per bank) SIMMs to fit the 32-bit bus width of the 486.
The bus speeds correspond to 25-120 MHz CPUs from Intel, ranging from 1x25 MHz to 3x40 MHz. The fastest chip available for these systems, possibly requiring a 5-to-3.3V adapter, was the 5x86 at 133 MHz (4x33). They were approximately equivalent to a 75 MHz Pentium.
Source: Scott Mueller's Upgrading and Repairing PCs, 11th ed. (August 1999)