Do not feed, poke or disturb the Itanic user
Privates on Parade
Intel Developer Forum All is well with IA-64, and to prove it Intel produced a panel full of happy Itanic users at IDF yesterday. It was a small room, and quite full, but the six guinea pigs were safe: members of the assembled press were only allowed to submit written questions on white pieces of card. These cards were handed from one PR to another, and finally placed before a panel moderator who decided if the panel could be exposed to the query.
"I can't ask that one!" said Meta Group moderator Phil Dawson at one stage, "that's too controversial!", so confirming the exquisite control-freakery of the occasion.
It's some time since Merced was passed over as a dud, and industry attention focussed on its successor. But it's the dead-end chip that Intel can't quite let go of. Although it's normal for microprocessor companies to discard early versions of a design that don't live up to expectations, with IA-64 Intel seems quite unable to declare closure.
So it's rather like watching a semiconductor Norman Bates who has lovingly preserved the processor in the fruit cellar downstairs. Intel feels obliged to have mock conversations with the dead chip. And it had one yesterday.
"Mother!" cries Bates. "What did you say Mother?"
Seven test representatives of Itanic test sites could answer. They were drawn from five technical scientific labs including NSCA, Caltech, Mayo Clinic and Ohio Supercomputer, plus Doubleclick, and each responded that it had been reliable and that they'd ported their applications without too many problems. The seventh member and token business user, and the only panel member using Itanic in a production environment, turned out to be Doug Busch, vice president of Information Technology at er, ... Intel. He wanted to run real business applications on Itanic, but admitted the closest he'd come was a proxy server. One day, he hoped, he could have Intel's entire ActiveDirectory loaded into memory on an Itanic cluster. But not yet.
"Mother! I've just cleaned the curtains!" cries Bates.
Checking our notes (March 28 last year to be precise), the talk about Itanium was all main-memory databases, business intelligence and decision support, and NUMA clusters. Yesterday, it was almost a full hour by our clock before a panel member volunteered any of these subjects. The tech compute guys were happy with floating point performance, for sure. And the cost? Mahlon Stacy of the Mayo Clinic said "We got it for free - so price/performance was excellent."
"She just goes a little mad sometimes. We all go a little mad sometimes. Haven't you?"
Two slightly risqué questions made it to the fruit cellar by proxy. Was 32bit performance as bad as all that? Intel's Mike Pope stepped in to answer on behalf of Ma Bates:- no one buys a 64bit chip for 32bit performance.
And did any of the panel prefer Itanic to the RISC chips they knew and loved? A terrible silence befell the panel. Tumbleweed blew across the room. Somewhere, we heard a church bell strike... And when someone finally spoke, it wasn't about RISC at all.
Strangely all this overshadows a strong IDF for Intel on the server side, and it hasn't even had its "server day" yet. Chipzilla trailed what should be a competitive chip with McKinley, introduced a new core P7 on a new process, and has Project Jackson multi-threaded SMT chips in the wings. And yet what we'll remember most of all is the unintentional Itanium comedy. Why the necrophilia? ®