David Kanter (dkanter@realworldtech.com) on 2/5/09 wrote:
---------------------------
>I just finished up a short article that discusses the delay in system shipments
>for Tukwila, the eagerly awaited Itanium that will modernize the system interface
>and bring Intel back to some semblance of competitiveness.

However no serious organisation EVER can buy new itanium supercomputer systems, simply because you do not know when intel delivers. Signing a deal a year and something in advance already is a huge advantage for a manufacturer, but you HAVE to deliver. With itanium, intel never could fulfil the most important condition that big organisations should demand from a manufacturer.

So the odds of new supercomputers getting bilt around an itanium core, tukwila or whatever its new name, would not be advisable. I would even call such commission: "not objective and ready to ship with retirement".

Because intel is that big, some years ago a lot of organisations betted that intel would deliver in time. They didn't with itanium.

Intel is doing the utmost minimum to keep itanium alive.
The only reason for sales for tukwila would be REPLACING existing itanium solutions, in order to upgrade them.

Most organisations by now already are doing business with a different intel department or others, especially IBM, for that.

Majority of itanium processors are IDLE. Only a few scientists who need big RAM sizes (which is very well possible with itanium) are interested in the platform and running some software at it.

The speed of the core is so disappointing after release of opteron and core2 (not to mention phenom2 and nehalem), thanks to the clock of the cpu.

You can't make up ipc wise in integers for a high clock times many cores. tukwila already in advance gets outgunned by more cores at a cheaper price by others.

You are comparing last itanium cpu's to Xeon MP. The last Xeon MP i can remember got released around 2003 or so?

The real platform to compare sales with is Xeon.

Beckton is the long awaited Xeon MP follow up from intel, which is real cool to toy with for many, yet companies massively will decide for 2 socket machines, simply because they deliver a lot of computational power already.

In all cases of course in HPC important is how many gflops something can deliver. Itanium can never compete there.

So its only left market is integers. I fail to see how it can compete with intels x64 cpu's, when using real world workloads.

Add to that, that itanium is known for using mighty amounts of power a core, which in todays society doesn't get seen as a big advantage. You can afford doing that only if you're the best. For something that is the fastest in some market segment, there is always takers, regardless of power usage.

With respect to future of course intel would only manage to position future itaniums if they are willing to sign a contract that at date Z they deliver cpu Itanium-X delivering this and that amount of gflops and integer achievement.

A vague statement of a 32 nm cpu in 2010 has no meaning. Tiny companies which produce something dirt cheap, cheaper than anyone else, they can do this. Intel can't afford itself doing that with itanium.

>Tukwila systems are now slated to ship in the middle of this year, as opposed to early this quarter.

>This article talks about the catalyst for the delays, which was primarily a change
>in the platform, and the history and decisions that ultimately lead to the delay.
>
>It also concludes with a look at the likely scheduling for the 32nm Itanium, code-named Poulson.
>
>Enjoy the reading,
>
>http://realworldtech.com/page.cfm?ArticleID=RWT020509000928