Kevin,
Here's an article on the next generation server bus fight. I bet AMD will be supporting Future I/O. Could be a big deal for AMD.
Bus war breaks out in PC server realm
By Rick Boyd-Merritt
EE Times
(01/08/99, 3:17 p.m. EDT)
SANTA CLARA, Calif. — A standoff over next-generation I/O schemes broke into a full-fledged bus war this week as Intel Corp. and three of its largest customers squared off with separate technology proposals, and one side lobbed the first threat of legal action. At stake is a strategic advantage in the PC server market, which has become one of the last bastions of significant profits and technical differentiation in mainstream computing.
Compaq Computer Corp., IBM Corp. and Hewlett-Packard Co. are expected to announce next week the formation of the Future I/O coalition to drive a new specification for a switch-based I/O architecture in high-end PC servers. A source close to the group said Compaq and IBM each will claim that a separate proposal from Intel, dubbed Next-Generation I/O, infringes on their intellectual property, opening the door to royalty claims against any of its adopters.
Intel dismissed the intellectual-property threat as saber rattling and said it lacked technical merit. Separately, Intel launched its own coalition around NGI/O this week, snagging five allies: Dell Computer Corp., Hitachi Ltd., NEC Corp., Siemens Information Communication Network Inc. and Sun Microsystems Inc. signed on as initial members of the NGI/O Industry Forum.
Both sides share a common goal: delivering a robust, high-capacity highway that links ever faster I/O subsystems to increasingly muscular processors. And both said they would negotiate any of their technical specs to reach a unifying deal on a single standard. But for now the two groups are locked in a high-stakes battle for the lucrative PC server turf, with both sides digging in their heels on the issue of intellectual property.
The Future I/O camp could release details of its technology, which grew out of their work on the PCI-X bus extension, as early as later this month. It is planning a developer's conference Feb. 11-12 in Monterey, Calif., aiming to steal thunder from Intel's planned NGI/O announcements at its own developer's forum in Palm Springs, starting Feb. 23.
"There will be a bus war," said Kimball Brown a chief server analyst for Dataquest Inc., a unit of Gartner Group Inc. A protracted stalemate could threaten Intel's business model, which is fragmented between Celeron CPUs that sell for as little as $70 at one extreme and high-margin Xeon chips with price tags up to $3,600 at the other.
Intel admits it has been only modestly successful in the former business and just started the server business in earnest last summer. "It's bad enough how they are doing in desktops, if they fail in servers they are really in trouble," said analyst Brown.
Intel needs to extend the I/O capabilities of its servers quickly, Brown said. Today's servers using PCI offer system throughput of about 500 Mbytes/second. But Intel's Foster processors, high-end 32-bit follow-ons to the Pentium II that are due out in 2000-2001, will require system throughput of as much as 3.2 Gbytes/s. And the 64-bit McKinley chips expected a year or two after that might take twice the bandwidth, he estimated.
"Intel's server business will stay mired in the low-end $10,000 boxes unless they can fix this I/O problem," Brown added. "Intel has a solution and just two years to get it out the door. They don't have time to work with a new industry group."
'More to lose'
Conversely, the Compaq-IBM-HP troika must defend its installed base of high-end servers. "These three guys have the most to lose from the PC horde," said Brown. Intellectual property is becoming a key sticking point and weapon in this dispute. Both sides say they will open their technologies for governance by an industry body and will license their specs on a royalty-free basis. But a Future I/O source said the group would not hesitate to seek royalties from NGI/O users.
"That's the first I've heard of that," said Bob Van Steenberg, general manager of enterprise severs at Dell, an NGI/O member. "They will have to prove that, and I would be surprised if it's true."
"They are trying to scare off the support from independent hardware vendors," said one NGI/O member. "We also have a lot of patents, and I would love to see what is in the Future I/O spec."
Though neither side could name at press time any third-party I/O companies as members of their respective groups, both claimed significant backing from card and chip makers. A Future I/O source claimed 3Com Corp., which owns as much as 50 percent of the network-card market, would be part of its coalition. For their part, NGI/O members noted that Intel and Sun both make I/O cards and companies such as Adaptec and Qlogic have worked on NGI/O.
With products not shipping for two years, Harry Mason, director of industry marketing at LSI Logic's Storage I/O Components Group (formerly Symbios Logic; Fort Collins, Colo.), said it is too early for I/O companies to take a stand for either side.
"We're not in a position to take sides," said Mason. "We certainly don't think it's going to stay divided, but it's going to take a while to settle out. Certainly there's a lot of infrastructure that needs to be built up before we can all throw away our PCI buses. We are just at the early stages of the evolution of that."
Indeed, some third-party I/O companies will become members of both groups and hold out hopes that the two will at least agree to some commonality at high layers of their architectures. "There are a number of people who hope there will be a convergence rather than a bus war," said Dell's Van Steenberg.
Tom McDonald, general manager of Intel's NGI/O division, a group the CPU maker officially christened just two months ago, dismissed charges that Intel's technology infringes patents of IBM or Compaq. "I question their motives," said McDonald. "This sounds like a lot of FUD [fear, uncertainty and doubt] to me. We might look at their IP and decide there is no merit to their claim of infringement. We haven't seen a bit of their spec, but our draft is publicly available from our Web site," he added. "They are developing this very much behind closed doors and have a business model which is all about trying to collect royalties from the computer industry. That's against what history has taught us about how to develop standards."
Tom Bradicich, director of architecture and technology for IBM's PC server group, would not confirm reports of the Future I/O plans, but he did suggest IBM was getting close to detailing some specific plans. "We are at a point where we want other people in the industry to know what we are doing," he said. "You are going to see significant support from the industry on this."
Still, he held open a door for an eleventh-hour deal with Intel. "There are some technical and business fundamentals that we are not willing to change. But I think we are pretty open to resolving technical differences," he said.
One senior engineer in the NGI/O group said his company will stand firm behind Intel. "This is what we are going to do, and we will put all our resources behind it."
The NGI/O draft that Intel has posted on its Web site shows an architecture that uses the basic physical-layer techniques common to Gigabit Ethernet and Fibre Channel to link the host with target channel adapters that interconnect across a generic switching fabric. Host and targets can remotely control each other's memories using a new protocol that's designed to enable end-to-end system latencies of a few hundred nanoseconds.
Unlike a PCI bus structure, NGI/O uses separate host and target channel adapters that use their own local memory pools, without slowing down a system CPU.
For its part, Future I/O had its genesis in PCI-X, another cooperative I/O effort by Compaq, IBM and HP. Targeted as a life extension for the PCI bus, PCI-X defines 64-bit versions at 66, 100 and 133 MHz using a new register-to-register data-transfer scheme. Aimed at systems that will ship later this year, PCI-X is wending its way through a working group at the PCI Special Interest Group.
— Additional reporting by David Lieberman
|
