How do PCI-E lanes work?

There are also 8 lanes coming off the chipset. That's where your network card comes from.

Note that the 980 would perform just as well on only 8 lanes.

as far as pcie lanes, it depends on your motherboard. You may have a pcie x16 (16 lanes) meaning that at least one slot is capable of pcie 16x. If they have two pcie slots, the specs will say something like pcie x16 or pcie x8/x8 or pcie x16/x8 . Meaning installing 1 card, it will use all 16 lanes. If it's x8/x8 that means it only has 16 lanes total and each card in dual configuration will use x8. If it says x16/x8 it means there are 24 lanes and the first card will use 16 while the second will only use 8. If it's 3 way sli or crossfire, then you might see something like x16/x8/x8 meaning the additional 2 cards will run at half the lanes as the first.

The chipset will allot the lanes to each device. Just the GPU plugged in will get all 16 lanes. Plug in others and lanes will be split between the devices. I don't think it is standard as to how they are split but dependent on the mobo.

Lots of debate out there but 8 lanes is plenty for any GPU on either 3.0 or 2.0 slot. This much most will agree on and I've never seen any proof to say otherwise.

I don't believe the chipset adds any lanes but controls the lanes the CPU provides. The CPU can handle 16 lanes and that's it. Whatever combination of devices has to share the 16 lanes. A pcie x1 slot will get its lanes whether it needs them all or not. Each slot gets its allotment and only it can use them while its still installed. The use is not dynamic based on need.

Some mobo's have an add on chip too add lanes. I'm not sure his this works but it kind of fakes it so the CPU thinks its only using its max 16 lanes. I'm not well versed on these but I believe this is the basic idea.

But if a slot is allotted 4 lanes it gets them no matter what. So your 16 lanes can be split as 8 for the GPU and 8 for the network card even though it does not need all 8 lanes. A third device will probably split the 8 again leaving the network card with 4 and the 3rd item with 4.

In the end since 8 lanes (and maybe even 4 3.0 lanes) is enough splitting it up and faking lanes really makes no difference to overall performance

Check out this article. It helps explain these extra chips that "add lanes" more than what the CPU has.

http://www.enthusiastpc.net/articles/00003/1.aspx

Its a good read and should help shed some light on things. Its very confusing I know and I'm not sure anyone truly understands it all

Perhaps give this a read, it explains the relationship with PCIe and graphics cards pretty well.
http://www.techpowerup.com/reviews/Intel/Ivy_Bridge_PCI-Express_Scaling/1.html

If your network card is in a x1 slot it is likely on a 2.0 lane, has nothing to do with the 3.0 lanes. Your 980 would run just fine on 8 lanes of 2.0, the link above will pretty much fill you in on bandwidth..

PCIe is rather complicated so I'll try to focus on answering just your question without getting into too much technical detail.

PCIe devices connect to the system through what's known as a root port, and each root port communicates with the system through a root complex. Each root complex controls one or more root ports.

Intel's CPUs that utilize the LGA-1156, LGA-1155, and LGA-1150 sockets have one 16x root port on the CPU. Intel's CPUs that utilize the LGA-2011 socket have three root ports on the CPU, two 16x and one 8x. These ports can be split into multiple sub ports in the following fashion.

Nehalem (excluding x58 based microprocessors), Westmere (excuding x58 based microprocessors, and Sandybridge: 16/0, or 8/8

Ivybridge, Haswell: 16/0/0, 8/8/0. 8/4/4

Sandybridge-E, Ivybridge-E, Haswell-E: 16/16/8/0/0/0/0/0/0/0 all the way down to 4/4/4/4/4/4/4/4/4/4 for up to 10 devices connected at once.

Each sub port can be individual down-negotiated, so it would be possible to configure an Ivybridge micrprocessor as 8/4/1 or 4/1/4. However, it is not possible to connect more than three devices to an Ivybridge or Haswell CPU without using an external PCIe switch, and it is not possible to connect more than two devices to Nehalem, Westmere, or Sandybridge.

In addition to the PCIe lanes exposed from the CPU itself, there is an additional 8x port on the PCH chipset. This port can be configured as 8/0/0/0/0/0/0/0 all the way down to 1/1/1/1/1/1/1/1 which is very ideal for connecting numerous low bandwidth add-in devices.
Most motherboards that include 4x and 1x slots connect those slots to these ports on the PCH, often with shared bandwidth. This enables the use of either a single 4x port in 4x mode, or a 4x port in 1x mode in addition to three 1x ports.

One of the beautiful aspects of PCIe is that it automatically negotiates link width. It is possible to insert a native 1x device into a full-speed 16x slot and have it run at 1x. The card will not be well secured but it will work. Furthermore, it is possible to cut the connector on a native 16x device (don't do this) and insert it into a 1x slot. If power requirements are satisfied, it will work at 1x link width. Similarly, placing tape over some of the connectors on a 16x device can reduce it to 8x, 4x, or 1x as appropriate.

Almost all graphics cards are native 16x devices but this is really just to take advantage of the 75 watts of power that 16x slots are required to deliver when requested. The bandwidth provided by a 16x slot is almost always sufficient; performance would only be impacted by utilizing a modern graphics card on a very old motherboard. For example, running a GTX 980 (native 16x PCIe 3.0) at 4x PCIe 1.1 may cause problems, but running it at 8x PCIe 3.0 will have a negligible impact on performance.