Next-Generation Intel Chipsets May Drop PCI Bus
As far as we're concerned, PCI can't die quickly enough. Intel began working on the standard 20 years ago and made it the preferred interconnect standard for second-generation Pentium systems as early as 1994. By 1999-2000, modern motherboards had all but dropped ISA (Industry Standard Architecture) and typically included just one slot per board. The then-new PCI standard supported a wide range of features ISA couldn't and offered 4x the bandwidth of the EISA bus.
The principle problem with PCI was its inability to scale either bit size or clockspeed. 64-bit PCI used a significantly larger physical slot, which made it more expensive, while the number of bus lines made increasing the base clockspeed tricky. Where consumer products were concerned, transfer speeds topped out at 133MB/s. This limitation led to the development of AGP, which was essentially a point-to-point PCI bus that supported additional features like sideband addressing and GART.
PCI Express x16 offers more than 6x the bandwidth of PCI-X, but uses a smaller slot
Modern devices—even PCI Express x1 devices—offer between 50-300 percent more bandwidth than PCI, depending on whether or not the PCI Express slot is a first or second-generation device. By the time PCI-Express 3.0 is finished, a single x1 connection will offer up to 800MB/s of bandwidth—6x more than first-generation PCI.
Backwards Compatibility Concerns:
The good news for consumers is that Intel's decision to drop PCI support should be all-but painless and may even improve peripheral performance in some cases. It's less common than it used to be, but some motherboard manufacturers still hang spare SATA/Ethernet/USB2 ports off the PCI bus, even though doing so adds latency and potentially reduces performance.
This decision may not sit well with users who rely on PCI slots for additional ports or video cards, but moving away from PCI is still a good idea, even if it penalizes small groups of people. The vast majority of add-in cards today are still built using PCI, but even a single mechanical hard drive can saturate its 133MB/s bus in certain circumstances. USB 3, next-generation FireWire, and RAID arrays are similarly pinched for bandwidth; we've reached the point where the PCI bus imposes a noticeable performance penalty on nearly every type of add-in card.
Unfortunately, the market as a whole is stuck in a reverse chicken/egg scenario. The majority of add-in cards are manufactured using PCI because it's the universal standard that everyone has and that people are most comfortable with. Because of this, manufacturers continue to use it, and thus perpetuate the cycle. Intel's clean break, when/if the company makes one, won't end PCI compatibility—third party manufacturers will continue to provide bridge chips—but it will send a signal to the computer industry as a whole that it's time to move on.
Next-generation products like SSDs, USB3, and even SATA 6G are only useful if their performance isn't handicapped by this sort of issue. If getting rid of PCI is the price we pay to ensure external devices are just as fast as their internal counterparts, we judge the trade well worth it.