The main downside to a serial protocol lies in the most simplistic difference between serial and parallel. In a parallel system, multiple data streams can be sent out and received, whereas with a serial protocol, data can only be sent out and received once per clock cycle. This drastically reduces the amount of data that can be sent and received, and also drastically lowers the bus utilization. There are other issues at the heart of a serial protocol dealing with the control signaling needing to be sent along the same line as data, but we won’t be covering it due to the relative complexity.

     When a series of PCI-Express lanes are combined, designers can easily up the amount of bandwidth that the connection is capable of handling. This fact can be easily seen in that there are currently five different slot sizes and specifications, ranging from x1 to x16, with x16 being the primary connection for graphics adapters since it offers the highest bandwidth. In total, there are 20 PCI-Express lanes available to the motherboard manufacturers. They can choose to use as many PCI-E connections as available, so long as they do not attempt to use more than the 20 lanes. On high performance motherboards this will likely lead to a single x16 slot, along with an arrangement of x1, x2, or one x4 slot.

     As with Intel’s new Direct Media Interface between the MCH and ICH, PCI-Express is a bi-directional, point-to-point bus. To put it in simpler terms, this means that the amount of bandwidth available to the interface is the same in both directions, as well as that multiple devices on the bus do not need to share the connection. This was a major drawback to the PCI interface, one reason behind the development of AGP some years ago. With PCI-Express Intel hopes to do away entirely with PCI, as well as AGP. With these new chipsets, Intel has finished off half the job, eliminating AGP support, though it will likely take some time longer to eliminate PCI. Many believe that AGP will be no more in roughly a year, and some estimates give regular PCI another two or three before it has been completely wiped from motherboards.

     Probably the most important issue that those looking to upgrade will want to look into is whether or not there will be any PCI-Express graphics cards available come time to upgrade. We know of several graphics manufacturers that are currently finishing off development of, or are ramping up for production of mid-range PCI-Express graphics cards. From the information we have, most of these first products will be based off of the mid-range chipsets, though high performance cards will also be available, only in more limited numbers. Both NVIDIA and ATI shipped out their high-end PCI-Express cards to reviewers for use in these Intel chipset reviews, so both are capable of providing working GPU’s right now, though they may need some additional tweaking before they arrive to the graphics manufacturers.

     The main difference between ATI and NVIDIA on the PCI-Express front is how they develop their GPU’s to deal with this new bus. NVIDIA has chosen to head down the route of developing a special interconnect bridge, HSI, for their PCI-Express enabled cards. NVIDIA will continue designing AGP based GPU’s and will then utilize the HSI Bridge to serve as a go-between for the PCI-Express and AGP interfaces. This is the most economical route as specialized GPU’s do not need to be developed for use with PCI-Express. ATI on the other hand, has chosen to implement a PCI-Express interface within the GPU itself. This will allow for maximum performance due to the native PCI-Express interface, but will force ATI to manufacturer far more GPU’s than they would should they have taken the same route as NVIDIA. ATI must design and manufacture specialized GPU’s for PCI-Express as well as AGP to supply their consumers. Both NVIDIA and ATI have chosen routes with pro’s and con’s, we’ll see who comes out on top in this bout between the graphics giants.