Thanks for the reference on Intel's Roadmap.
My interpretation of the roadmap is that I2O will appear on 450NX chipsets for the Deschutes (Pentium II) processor for servers by the first half of 1998. I2O also will be integrated into the high-end version of the 440BX chipset needed for Pentium II PCs running 350 MHz and up clock rates, which will begin appearing during the second half of 1998.
This suggests that by 1999, most Intel servers will have I2O capability on the motherboard, as will most high-end PC's. As a matter of fact, by then the only high-end PCs that will not benefit from I2O on the motherboard would be those that use the PCI to ISA Xelerator version of the 440BX, which appears not to be in the mainstream direction of things to come, since it will support neither `1394' nor the multi-processor extension.
Am I correct in assuming that the PCI to ISA Xelerator is a bone to legacy attachments that must run ISA-bus devices into a speed-demon computer. In time, wouldn't most, if not all, PCs gravitate toward the high-end solution incorporating `1394' for fast serial devices and I2O for multi-processing if needed and/or for lots of other, advanced I/O?
Boiling all this down, it appears that I2O on the motherboard will be a reality during 1998, both for servers and PCs. Intel-Architecture (I-A) servers should be in the mid single-digit millions during 1999. High-end PCs sporting I2O on the motherboard should at least equal the number of servers in 1999.
By 2000, virtually all servers, I-A and others (e.g. DEC-architecture, HP-architecture, if not IBM and Sun), should implement I2O on the motherboard, probably totaling high single-digit millions of units.
But also by 2000, I2O on the motherboard of PCs should be commonplace, say 50%, which should equate to 40 million to 50 million units.
Meanwhile, I2O will be found starting now on NICs, RAID devices, network hard drives, network backup systems, and network devices of all types, including hubs, routers, switches, etc. According to DEC, expect I2O even on NCs. High-end servers can also benefit from multiple instances of integrated I2O chips, numbering easily in the double digits per server.
The number of devices times the market penetration expected of each type of device suggests that these uses of I2O will begin to ramp up significantly toward the end of this year, amounting always to many more units than the number of servers alone. For the remainder of this year, and next, these uses of I2O will dominate in numbers. By 1999, the number installed on motherboards should begin to ramp rapidly. By 2000, network devices and the PC motherboard numbers both should be major drivers of I2O usage.
As I2O becomes the de facto open standard, the benefits of standardization, coupled with cost reductions, should propel I2O usage to being compulsory for all networked devices and perhaps even for stand-alone devices, as traditional drivers start giving way to I2O-based configuration for any add-in I/O capability. When this happens, the numbers should really sky-rocket to being far in excess of 100 million units annually.
However, being the staunch conservative that I am, for now I am staying with my current, lower estimates. These are 1.9 million units in FY 1998, 9 million units in FY 1999, 22 million units in FY 2000, and 53 million units in FY 2001.
Forget the limitless, unending stream of royalty revenue for a moment. As the 21st Century unfolds, virtually all I/O throughout the whole world will be handled by WIND's product. By the year 2001, there won't be a single person, nor a communicating device, that each day doesn't touch this one product from WIND. How many other companies in your portfolio are slated to have that kind of impact on humankind?
Allen |
