Intel gets off the bus--déjà vu?
A long time ago (October 2000), in a galaxy far, far away, InfiniBand came into being as a concept for the way hardware devices should communicate.
PCI--the existing standard--is like driving a golf cart at the Indianapolis 500. With bandwidth from 1Gb/s for 32-bit 33MHz slots to 4Gb/s for 64-bit 66MHz slots, PCI presents a shared bandwidth with each device connected to it reserving a chunk of the bandwidth for itself. Founded by Compaq, Dell, Hewlett-Packard, IBM, Intel, Microsoft and Sun Microsystems, and now supported by more than 70 additional companies, InfiniBand was intended to be the perfect mix of speed and cost effectiveness--an architecture that would unite the IT world with speeds starting at 2.5Gb/s with a 1x connection up to 12x (30Gb/s).
Why did Intel jump ship? To be fair, Intel says it still supports InfiniBand. However, it's been developing its own I/O standard, 3GIO--or, as it's now known, PCI Express. Naturally, the mavens of computing insist that PCI Express is not a rival candidate to InfiniBand. It's an adjunct, a complementary technology, they say. But let's think back to two bus architectures of long ago, ISA and VESA Local Bus (VL-Bus).
VL-Bus arrived as a compliment to ISA--a response to the need for more graphics speed. It worked well and was on its way to a second version, backed by Intel, among others, when suddenly Intel withdrew, as it had a specification of its own: PCI. It was still in rough form and hadn't been practically implemented anywhere, but PCI--not VL-Bus--was Intel's vision of the future. When was the last time you saw a motherboard with a VL-Bus slot? PCI became not just an adjunct to ISA, it replaced it entirely. We're forming a picture here.
Intel found itself in the position of being unable to control the direction that VESA was taking with its local bus architecture. Now remember that Intel signed on for InfiniBand 1x--not 4x or 12x, the speeds emphasized by the rest of the association. Once again Intel was not in control, and once again Intel backed away from the specification. Do you think the pundits are correct in claiming that PCI Express and InfiniBand will live together in harmony? Can we repeal the laws of inertia?
Despite the fact that InfiniBand will arrive (in a practical fashion) at least 6 months (and possibly a year by some accounts) before PCI Express, don't think this is going to be a bloodless coup. Now that Intel has withdrawn, IBM is one of the biggest backers of InfiniBand. IBM is also a part of the PCI Special Interest Group (PCI-SIG). It hasn't withdrawn from that position. IBM and Intel are best of friends, aren't they? It's a big picture.
What's your take on the future of InfiniBand? TalkBack to us or send e-mail to Bill.