Why do graphics card power cables (PCIe, etc) use more than two wires?

Question

High-end graphics cards need more power than can be delivered through the PCI slot. One option would have been to deliver the extra power using a 4-pin Molex or SATA power connector, but recent graphics cards need even more power than these can deliver.

The computer industry has introduced a series of "PCIe power connectors", each having two rows of pins: one for +12V and one for ground. The later connectors add more pins (4->6->8) and more wires to carry more current.

Question: why did the industry decide to add more pins instead of using thicker wires? In particular, is there an electrical safety/reliability reason for using a larger number of low-current conductors instead of a smaller number of high-current conductors?

Adding twice as many pins bumps the connector and cable price up much more than using pins and wires that are rated for twice the current (up to a point of course, but PCIe is nowhere near the point of diminishing returns). In theory you can use a pair of 4-pin wires in place of a single 8-pin wire, but in practice this doesn't really happen all that often and I doubt it was the deciding factor.

I can certainly see how manufacturers want to make sure users don't use a low-current cable on a high-current connectors, so if they had stuck with 2-conductor wiring I'm sure they wouldn't have left the connector unchanged. Most likely they would have adjusted the keying on the socket (that's one great thing about those Molex mini-fit connectors, they've got an exponentially large space of keying choices to stop people from doing stupid things).

Answer 1

Modularity

As @helloworld922 and @Ignacio Vazquez-Abrams point out, a thick wire is extremely difficult to handle in the tight confines of a computer chassis (which are getting ever smaller to meet consumer preferences). The power connector is attached to a card which must exit the bend radius within the specified spacing to the adjacent card so there is limited ability to just make the wire thicker. Further, accommodating thicker wires presents fewer options for placement and routing of the PCB making design of the PCB and guaranteeing sufficient airflow around the cable and connector system more difficult.

By using multiple connection points, the spec can accommodate future reductions in power demand or differences between high-end cards with very large requirements and smaller/low-end cards with less. This is especially important in the modern computing era where performance comes from parallelization (more cores, rather than higher performing ones).

Consumer Safety/Confusion

@Tom Carpenter writes, "That and if they used the same connector but with thicker wires, someone would inevitably have an old model supply with thinner wires and use that for a new card without stopping to check. An important lesson in engineering is: Assume the user is an idiot. Many won't be, but there is always one!" -- It's apt advice. Further, many users would be confused by the subtle distinctions.

Cost

Cost is not obvious, but it is likely much cheaper to implement it the way it has been done in the PCIe spec. The majority of the cost in short-run cable assemblies is actually the manufacture of the terminations (end connectors) and the assembly of these terminations to the wire -- not the wire itself. Thicker cables have much lower yields when using crimp or insulation displacement manufacturing techniques and the larger cables necessary to accommodate the PCIe power goals would likely not fit in the existing tooling and machines used to assemble the molex, audio, and other PC wiring. PC cabling is manufactured in such large volumes that even small differences in yield make a substantial difference.