The PCI-X specification, as the initiative is called, reflects a growing frustration among server vendors with the slow pace of innovation around how data moves inside a computer, according to sources. Intel boosts the speed of microprocessors on roughly a quarterly basis. However, when it comes to boosting the speed at which data can be shuttled around from the microprocessor to other components in a computer, change has been glacial.
In fact, the chip giant has resisted changes in the current architecture, which it developed, and is said to be working on its own proposal.
The new PCI-X specification would quadruple the speed of the system bus--or the data path between the processor and the rest of the computer--to 133 MHz. The processor connects to the hard drive, networking cards, and other computer components through the PCI (Peripheral Component Interconnect) bus. The fastest PCI bus currently runs at 66 MHz and can pass up to 64 bits of information per clock cycle.
However, these are recent innovations. The vast majority of desktop computers and lower-end and midrange servers continue to use the 33-MHz PCI bus with a 32-bit gate developed by Intel that was released in 1993.
The PCI-X specification proposes a 133-MHz PCI bus that can handle up to 64 bits of data per clock cycle, sources close to the companies said. In addition, Compaq, IBM, and HP have presented the specification to the PCI Special Interest Group, which sets the standards for PCI technologies.00000
Something like the PCI-X specification is long overdue, according to Peter Glaskowsky, an analyst with MicroDesign Resources.
"Intel is late in getting a successor to the PCI bus. It is almost full and they need to introduce something fast," he said. "The bottleneck right now is the system bus." Intel has been working on a successor to the 33-MHz bus, but few details have emerged, Glaskowsky added.
The circumstances surrounding the development of the proposal seem to indicate that tensions between Intel and the computer server vendors exist. An Intel spokesman admitted that the company only learned of the proposal recently. "We have it now and we are studying it," he said.
The spokesman further pointed out that server vendors and Intel differed on the boost of the PCI bus to 66 MHz. Intel resisted the change, but the server vendors pushed for it.
Although differences exist, Glaskowsky said that cooperation will be the likely result. The PCI bus is embedded in the chipset and Intel is the largest vendor of chipsets. Therefore, its support is necessary.
At the same time, kowtowing to the server vendors will likely not take away revenue opportunities for Intel. PCI technology has become a standard throughout the industry because it comes fairly cheap. Members of the special interest group get a license for PCI technology royalty-free. Right now, Intel does not gain royalties from PCI specifications submitted to the group. The PCI-X proponents would not earn royalties either.
Companies have tried to charge for bus technology in the past. IBM, for instance, tried to popularize a technology they called Micro Channel in the early 90s. Micro Channel came with royalty fees and it failed.
Some have theorized that Intel is challenging the PCI-X specification because it plans to change royalties on its own standard. Such a move, however, would likely fail. Some of the vendors that would have to license such a technology are making parts that sell for $5, according to Glaskowsky. A licensing fee would be prohibitive.
"If Intel is thinking of doing something proprietary, the effort is doomed," he said.
The Intel spokesman added that the company often provides technological standards free of charge to gain industry acceptance. Intel, for example, gives the intellectual property underlying the AGP (Accelerated Graphics Port), a bus for graphics chips the company released last year, royalty-free to vendors.
Intel is an investor in CNET: The Computer Network.