Amid CPU hype, I/O is key at IDF

According to Intel, "Highlights [of the Intel Developer Forum] include keynote speeches from Intel top executives providing vision and technology direction; nearly 200 hours of lectures, sessions and labs; and a technology showcase featuring new technologies, developments and initiatives from established and emerging companies and industry groups. The theme for IDF Spring 2002 is 'Advancing the Digital Universe.'"

Let's put aside the fact that it's not Spring and that the universe in which we live is decidedly analog in nature. We can forgive Intel a bit of hyperbole in some things. The real focus of this latest IDF will, most likely, be speed. And while CPU clock speed is always a topic that generates energetic conversation, this year we might see other candidates take center stage: 802.11, optical networking, InfiniBand, Ultra-wideband RF, Gigabit Ethernet, and 3GIO. The truth is that we actually have enough CPU speed, in fact more than enough. We've also been addressing external connectivity speeds over the course of the last two years. What we're missing is the I/O speed necessary to transfer data to and from the CPU and then into the outside world. It's like speeding around between cities in a fast sports car only to switch to a golf cart once you're inside the city limits.

Intel's plan to alleviate the huge speed differential includes a new 3rd Generation Input/Output bus architecture called 3GIO for now although it may eventually be monikered as PCI 3.0. It offers a 6x increase in speed over PCI-X's 1.1GB/s rate and that means faster data handling of all types--anything that travels up and down the bus, anything heading for the outside world--and it should keep us going for maybe the next decade, says Intel. No less than <="" span="">f="http://techupdate.cnet.com/enterprise/0-9500-721-235085.html" target="_blank">Dell agrees, and Joe Sekel, senior server architect and technologist for Dell's Enterprise Systems Group is talking up 3GIO as the near perfect match for InfiniBand.

Essentially he's correct but remember, this is an Intel  Developer Forum. What you may not hear is that 3GIO probably won't appear in any form before Q103. With most manufacturers already lining up their '03 products, that may make any implementation of 3GIO before Q3/Q403 an impossibility. You also may not hear that Intel's track record with external bus architectures has been abysmal. The first implementation of PCI was, at the very least, buggy. The first appearance of the AGP bus drove graphics card manufacturers to avoid using its theoretically higher speed as often as possible. Given past performance, even the most optimistic of Vegas oddsmaker might set Q1/Q204 as the date for the first real appearance of a functional 3GIO bus.

And because this is an Intel forum, it's unlikely you'll hear about the current competition to 3GIO, called HyperTransport, from AMD. HyperTransport is a technology that's designed to "allow chips inside of PCs, networking and communications devices to communicate with each other up to 48 times faster than with some existing bus technologies." The reality of this architecture is that AMD's 8000 series chipset, which supports HyperTransport and dovetails it with AMD's 32/64-bit Hammer CPU, will be available by Q402. That could put HyperTransport in the box by Q1/Q203--well ahead of Intel's 3GIO. "Could" is a good word to use here because Intel has a way of winning out no matter what. You won't find a VL-Bus slot in any computer built since Intel released AGP and even ISA has moved from rare to extinct in the face of PCI.

The problem is, lower prices and superior product are the natural results of competition. So far, for whatever reason, the industry has managed to downplay the evolutionary significance of AMD's Hammer and you can already hear the first sideswipes at HyperTransport beginning. You can't afford to let that happen. Your IT dollars are shrinking and anything that will decrease your costs helps keep your TCO healthy. From your letters, it's apparent that AMD has achieved a significant foothold in the desktop workstation arena because of the lower CPU pricing and resulting lower workstations cost. There is no evidence, other than the usual whispered hearsay, that it can't do the same for your servers.

What's your take on 3GIO? Will it be the next iteration of PCI? Post your thoughts in our Talkback forum or drop a line to Bill.