Leonard Tsai, chief technologist at NEC, made some interesting comments regarding Intel's Itanium processors at a conference back in July. Since then he has departed from the firm, "under good will", though disagreeing with some of his colleagues about some aspects of Itanium.
Tsai observed that the new Explicitly Parallel Instruction Computing (Epic) instruction set associated with the Itanium processors could take years to learn and exploit properly. I would be surprised if he was wrong.
Is he correct and, if he is, what does it mean for the future of server systems?
Well, it would be very surprising if Intel had managed to develop a brand-new instruction set that every applications developer in the world could use without the need for training. Of course, the available development tools would allow most of them to get it to work - as in carry out a task reasonably well. But getting an Itanium server to fulfil its potential may take some time.
Is that a problem for the majority of users? Perhaps. Those of us with long memories will remember the last major change in Intel's architecture - from 16bit 8086 and 286 processors, to the 32bits of the 386. That was back in 1985, by the way. The official operating system of choice for that processor was, in fact, Unix. But what happened? There were no applications available, particularly for Unix, because developers had to learn a whole new instruction set.
So the old Microsoft applications for 16bit systems won the new 386 marketplace by default. The fact that they were available provided Intel with a market to exploit. It was several years before native 32bit applications started to arrive on the scene.
I can see no compelling reason why progress should be any faster this time, although Intel may want to hurry things along because it has bet so much development money on Itanium, and wants to get some of it back before the gamblers of the US stock market make a total nonsense of every IT-related business.
Things may move more quickly if Intel guides the chip towards a new breed of plug-and-play server. Itanium offers the tantalising prospect of servers-as-commodity. Can't get what you need from IBM today? Then buy it from Dell, HP, NEC, or pick another brand. Swapping servers to run the applications and services may become simple and straightforward.
The attraction of such servers won't be in the Intel hardware but in the software and services. And getting the most from the hardware by fully exploiting the new instruction set will be the real market of the future.
Epic will probably take some learning but it is likely to contain some really useful tools for developers to exploit. Intel's own history has clearly shown that there is no shame in admitting this to be the case.
Have your say: contact IT Week
reader comments