Intel plots post-PCI, 3G PC standard
Convergence, here we come...
Intel Developer Forum Intel Developer Forum, San Jose So, farewell then, PCI... So, farewell then, AGP... Intel is already working on your successor, a unified PC sub-system designed to provide connectivity out into the next decade of the 21st Century.
As yet, this third-generation technology has no name - at least, not one that Louis Burns, Intel's desktop products chief was willing to reveal at its announcement - but Chipzilla is already promising a preliminary spec. next autumn.
By then, the company reckons it will be able to tell the PC industry how it will be able to provide a full serial, point-to-point interconnect that's capable of running at clock rates of 10GHz and beyond to match the ramp of CPU clocks speeds over the next decade. It also has to be able to surpass the limitations of today's copper connections by utilising optical links.
Those were the key requirements Burns outlined this morning. The goal, he said, is to develop an I/O system capable of holding its own for the next ten years. Other would-be successors to the now well established PCI bus - including not only AMD's recently announced HyperTransport (aka Lightning Data Transport) and, interestingly, AMD's own AGP system - just won't cut it, he claimed.
Many are good solutions to some of the limitations of PCI, he suggested, "but they don't have the legs to run for the next ten years."
Sustainability over time will be key to persuading the industry that Intel's solution - a solution that doesn't even exist yet as a specification - is the one to follow. To tempt those who might be more keen to follow HyperTransport and PCI-X, both of exist as something more than a concept and a technology evaluation programme, Burns promised Intel would establish its system as "a public standard", surrendering its IP to the wider industry, we gather.
Of course, that doesn't preclude the kind of industry conflict that emerged when the old ISA bus evolved into Extended ISA (EISA) alongside IBM's alternative bus, Micro Channel Architecture (MCA), and it's not hard to see battle lines being drawn with AMD's HyperTransport supporters on one side and Intel's third-gen. gang on the other.
The scheme here is gear the PC up to support the kind of rich media content broadband networking is going start delivering in the not-so-distant future. The Pentium 4 is the core of Intel's digital convergence solution, and the successor to PCI is all about allowing the rest of the system to cope with the vast amounts of data involved. Compression will only take you so far, particularly with multiple sources being accessed simultaneously.
It's not needed yet, which is why Intel is looking ten year down the line and planning to integrate native wireless and USB 2.0 support into its chipsets in the meantime as an interim measure (see Intel to integrate USB, wireless into P4 chipsets).
The other side of the equation is, of course, the InfiniBand infrastructure running, ideally (in Intel's mind, at any rate), on Intanium hardware. Together they will store, process and transmit all this digital data that the P4 family is designed to work on at the client end - be they PCs, set-tops or future generations of X-box - and present to the user.
The essential point here, though, is that a solid successor to PCI should ensure that the general purpose PC remains able to compete with hugely powerful convergence systems like, not to put too fine a point on it, X-box and thin clients connected to InfiniBand-connected mega-servers.
The most consistently asked question at IDF is: what do users need the power of a P4 for? Intel shrugs the question off, pointing out that everyone asked the same thing during the 286 to 386 transition, and every other shift to a new CPU architecture since then. The difference - which Intel doesn't mention - is that this time there are other machines competing for consumers' disposable income. That's why the question has new relevance, and why Intel needs to start finding answers.
It's Extended PC Era marketecture is one (also currently being offered up by Apple, you'll note), so is the PCI follow up. We'll tell you in ten years if it's worked. ®