For various reasons - historic, technical and marketing - there is not just a single type of socket on a motherboard which can accept any x86 CPU from any manufacturer. In fact, there are three sorts of slot and at least a dozen sorts of socket (two of the slots and two of the sockets are visually indistinguishable, to avoid inadvertent convenience) at the moment.
This socket was used by the earliest Intel Pentium chips. It has 273 pins.
This socket is for later Intel Pentium chips, and has 320 pins.
This socket was introduced for the Pentium MMX, though it can take all the chips from the P75 upwards. It is used for essentially all the Pentium clones with a 66MHz bus. It has 321 pins, and provision for supplying the split core/IO voltage required for Pentium MMX and later chips
The Pentium clone manufacturers are unable to use Slot 1, since Intel holds patents on it. They therefore agreed an advanced version of Socket 7 motherboards, using the same socket but running the bus at 100MHz; the chipsets used provide AGP, which Intel was keeping as an argument for Slot 1. These are called Super 7 boards.
This socket was used for the Intel Pentium Pro chip, and essentially nothing else (except the PPro Overdrive). It is rectangular, and has 370 pins.
The reason for the rectangular socket is the rectangular chip that plugs into it; the reason for the rectangular chip is that it contains two (or, in the case of the rare and incredibly expensive PPro with 1M cache, three) separate dice. Yes, Socket 8 chips have level-2 cache in the package, though not on the core - the ceramic package contains a very small circuit board with the two or three dice mounted directly onto it.
Unsurprisingly, this was extremely expensive for Intel to manufacture - if you damaged any of the dice in the bonding process, or if there was a defect in the in-module circuit board, the module was useless - so they moved to Slot 1 as soon as they reasonably could.
Socket 8 has enough pins to run a multi-processor protocol enabling you to put up to four CPUs in a system without additional effort from the chipset.
This is not a socket, but a slot - processors for Slot 1 come in a black metal cartridge, containing a circuit board with several packaged chips mounted on it, which plugs into a 242-way connector on the motherboard. The circuit board inside the package can have 512k of cache on it, consisting of two 256k cache chips; this cache runs at half the CPU speed. Alternatively, if the core being packaged is Coppermine, there is no cache on the board; the Slot 1 package is used for compatibility reasons
For modules containing Intel's extremely fast SRAM chips, Slot 1 is not sufficient; it's not possible to cool the SRAM chips as well as the core. Accordingly, Intel introduced Slot 2, which holds much larger cartridges than Slot 1; Slot 2 processors have cache running at the full CPU speed, and can have 512k, 1024k or 2048k of the stuff. The slot has more connectors than Slot 1, to support a more aggressive multi-processor protocol amongst other features.
The electrical design for Slot 2 was changed significantly with the advent of Coppermine and Cascades; the new cartridges look the same as the old ones, but have in-built voltage regulators.
The slot-based processors come in cartridges; these are much cheaper than Socket 8-style MCMs if you have more than one die to package, but are expensive overkill for a single-die chip. Accordingly, at the start of 1999, Intel moved its Celeron range of processors to a new socket, Socket 370. Chips for this socket contain a single die, with L2 cache integrated, and are sold very cheaply.
This is a format very similar to Socket 370 (it has 370 pins, but rather different voltage requirements to the original Socket 370), in which certain Coppermine chips are packaged. The novel feature is that the processor die is mounted directly on the surface of the substrate, so that the heat sink may be attached to its back for optimal heat conduction.
The P3S processor, and Celerons above 1GHz, use the Tualatin core, and are packaged in FCPGA2, which adds an Integral Heat Spreader (see Socket 423) to the FCPGA design.
Willamette's totally new front-side bus means that Intel needed to introduce yet another form factor to account for the extra pins; hence the Socket 423. A distinguishing feature of the packaging is the Integral Heat Spreader (IHS), the large metal slug under which the die is hidden; this provides a large, robust and well-conducting surface to which big heat sinks can be attached.
About ten months after introducing the Pentium 4, and simultaneous with the release of the 1.9 and 2GHz models, Intel redesigned the packaging entirely. The new package is substantially smaller -- not much larger than the IHS -- whilst being equipped with more power pins. I have not yet encountered a satisfactory explanation for this change; it may simply be to enable smaller form factor P4 designs, or may be to address power-supply problems that cropped up when the 0.13u CPUs were being tested.
The Banias chips, launched as Pentium M in March 2003, use a socket with 478 pins, but a slightly different arrangement (two pins in one corner are moved) so the chips don't fit into the normal 478-pin socket. I don't know why this was done; they'd be superb chips for a low-power silent desktop.
Socket 603 is the connector for Intel's Xeon-branded server Pentium4 processors. The additional pins are for providing more power to future CPUs with large on-die level-three caches, and for providing some inter-processor-communication signals for systems with more than two CPUs
Athlon chips use a connector identical to Slot 1 (to take advantage of the large number of companies that make such connectors), but speak a completely different protocol across the connector.
In June 2000, AMD released its first Athlon processors with on-die L2 cache; since there is only one chip to package for these CPUs, they have moved from Slot A to the cheaper-to-manufacture 462-pin Socket A, which is now used throughout their range. Changes in FSB speed and power and thermal requirements mean it hasn't been possible to plug a early-2003 CPU into a late-2000 motherboard, so the upgrade path isn't completely trivial, but the reverse is still possible.
The Opteron uses a new socket with many pins, for handling its rather substantial power requirements, and the very large number of connections required for its two DDR333 memory channels.