In-Depth

Q&A: IBM’s Itanium Plans

Big Blue announces two new Itanium 2-based systems in just six months

• By Stephen Swoyer
• 12/02/2003

We spoke recently with Jay Bretzmann, IBM manager of eServer products, about Big Blue’s new Itanium server. We asked Bretzmann about his company’s commitment to Itanium—hint: he doesn’t believe Big Blue has been ambivalent about the chip. We also prevailed upon him to explain how IBM positions Itanium relative to its homegrown Power chip, as well as AMD’s Opteron, a new dark horse entrant in the 64-bit chip wars.

You announced the xSeries x455 a new Itanium-based system [last month], which is actually your second Itanium system since this spring. How are you positioning this system relative to the [xSeries x450] Itanium system that you announced this spring?

We actually think that the sweet spot for Itanium 2 is in the four- to eight-way space, and we have what we call a pay-as-you-go value proposition. That’s where we say that we’ve got a better way for customers to get to 16-way SMP, a way that saves them money upfront, and that’s scalable to suit their needs. So back in July, we introduced a four-way [x450], because that’s where we feel most of the demand is right now.

Is that expandable to 16-way?

No, that’s four-way only. The x455 [which IBM announced last month] is the scalable version of our Itanium product line, and it’s based on the second generation of our Enterprise X Architecture [EXA, IBM’s SMP chipset for Intel’s Xeon and Itanium chips]. It’s a four-way [system] that you can build in blocks [of other four-way x455 systems] to get to 16-way, so it truly is pay-as-you-go. It gives [customers] a lower entry point [to get to Itanium] without the penalty to swap out and change the server if their needs grow.

You’ve announced that you’re going to take your 16-way Xeon-based system [xSeries x445] up to 32-way. Do you plan to do the same thing with the x455?

We’ve had a little confusion between the 455 and the 445, which is a 32-way Xeon[-based] system. With the 445, we have a product plan saying we’re going to 32-way, but we have not yet made any similar decisions about the 455, although the [EXA] chipset architecture would support it. One big difference here is density, and that has to do with the power consumed by the processor. In the 32-bit space, we can pack eight [Xeon microprocessors] in a 4U chassis. Now, with the Itanium2, because they’ve got all that cache and they’re consuming 130 watts apiece, we get half density, we only get four.

You did say that it’s possible to get to 32-way, right? There’ve been rumors that you plan to introduce a 32-way Itanium-based system, probably next year…?

We haven’t made that judgment yet. The chipset architecture does support it, so there’s no way we couldn’t do it. But another thing that you need to consider is that it’s a very big test build. When you put together a test configuration, especially with the Remote I/O [a separate I/O chassis that provides I/O expansion for IBM servers], you’re spending millions of dollars buying the equipment and testing it to make sure that everything works to IBM’s satisfaction. So that’s a real limiting factor. If I thought that there would be considerable demand for a 32-way Itanium 2 system, I’d make the call and say let’s spend the money, but you’ve got to walk until you run and we’re gong to continue to evaluate the need, because we don’t believe that there’s that kind of demand yet.

Okay—you’ve obviously been less enthusiastic about Itanium than some of your competitors, such as, say, HP. Some folks have even accused you of being downright ambivalent about the chip. Since April, however, you’ve made two [Itanium] product announcements. What kind of commitment would you say that IBM has to Intel’s flagship 64-bit chip?

Well, I don’t think we’ve been ambivalent at all. I think we’ve kind of timed our introduction, so, yes, one of our competitors was out there, say, nine months before we were. But they had an operating system [HP-UX] that ran on it, while we were waiting for Windows. If you precede the market by a long shot, you don’t’ really have an opportunity to sell too many of them, so we started out in this marketplace right after Windows Server 2003 came out. At the time we were shipping, the McKinley erratum popped up [a flaw in the Itanium 2 “McKinley” design], and due to IBM’s policy of zero tolerance of lack of integrity, we took those products out of market. We knew that Madison was coming, so we just basically delayed until the Madison time frame.

How do you position Itanium relative to Power, IBM’s homegrown 64-bit architecture? Do you see the two as on equal footing, then?

The Power architecture is a much more mature core than Itanium. We’ve been building Power systems for a long time, so we actually ran a multicore design already, I believe first in the 64-bit space, and we’ll see additional multicore designs. A lot of it is a matter of how long you’ve been building the circuits, because the first thing you do is design the logic, then a shrink process to shorten the circuits, then a design process, and the fact is that we’ve been doing that longer [with Power] than Intel [with Itanium]. So I wouldn’t’ say that Itanium will always be number two, because each product has its own strengths.

Such as?

One of the big differences with Itanium 2 is that it really does have great floating-point performance. It’s a brand new floating-point core, so it really screams when it comes to floating-point performance.

It’s superior to the floating-point performance of Power?

Yes. Itanium 2 has superior floating-point capabilities. It’s a brand new design and has multiple floating-point units on the chip. That’s what gives it the edge.

There’s a dark horse in the 64-bit race these days, AMD’s Opteron chip. IBM has already announced a commitment to Opteron, of course, but I’m wondering how you’d position this 64-bit contender relative to Itanium and Power?

To date, what we’ve seen is the request to ship a system that’s appropriate for each market, so we put together a two-way system and we sell both of those in heavily clustered configurations. We continue to monitor demand for a larger system built on an Opteron processor, but I have no plans at this point in time to introduce one. We believe in offering our customer the choice, so it doesn’t matter what our customers need, we can sell it to them, the 64-bit addressing version, the 64-bit addressing version of Intel-compatible design, the fully 64-bit power architecture in the RISC space, we think that with all of those processors…

Where do you see most of the demand for Itanium coming from? Will most of the uptake be on Windows, on Linux, or split pretty evenly between both?

It sort of divides into two camps, I would think there would probably be a little more database-serving interest on Windows to begin with, because there are a lot of Windows customers looking for SQL Server or large SQL Server databases. For something like SAP, it’s kind of a mix of both. The people that deploy SAP earlier usually are on Windows, the people who are rolling it out later are increasingly looking for Linux.