This morning, the Itanium Solutions Alliance put out a statement with some vague numbers showing the strength of the Itanium server platform in 2008. By IDC’s reckoning of the server space in the fourth quarter, shipments of Itanium-based machines rose by 18 per cent and it was the seventh straight quarter of sales that crested above $1bn for the Itanium server category. Data from Gartner’s report covering 2008’s server sales indicated that Itanium machines outgrew RISC-based alternatives in terms of sales and shipments, growing share in each category.
Unlike the majority of people I like Itanium (especially the new version comming out, its sooo goood. Not as good as SUN’s ROCK, but good).
Itanium runs the 3 OS’s I like most (QNX aside); OpenVMS, Windows server 2008 R2, and Red Hat linux. It is a good archetecture and the new revision is really going to show it off in good light (finally). Good luck Intel, you need it here.
I was just thinking when I saw the headline whether we’ll see in the future Solaris for Itanium and Solaris for POWER along with Solaris for x86/x64 and SPARC. There was an ‘early access’ version available for Itanium of Solaris, so theoretically there is no reason why they couldn’t do it.
What I can see is Oracle being able to provide one operating system spanned across multiple platforms, and being able to offer their product on all those said platforms.
I really like Itanium as well. I’ve got a 733 Itanium 1 and a 1.4 Itanium 2. I’d love to get my hands on one of the newest models but no one seems to build anything in my price range so I have to wait until they show up on eBay.
So, do you use one?
Not in an academic, investigative or toy way – do you use one as your primary machine?
Itanium looks great on paper. The practice is rather more sober. It loses on price/performance, it loses on performance/watt, and it loses on compatibility. Unless you need the scalability it offers on the very high end, it is just not that interesting. As the Amd64 compatible chips move upmarket, it will be less interesting there too.
There is a good parallel with the Alpha. It was also a good chip for a time, but couldn’t really compete with commodity hardware. Having Windows support is nice, but limited without any Windows applications. When Alpha died, many geeks (including me) lamented, but the market barely noticed. It just didn’t reach the mass necessary to sustain an architecture.
The alpha died not because of market loss, but because of acquisition after acquisition.
The acquiring company Compaq, wanted to make an entrance into the high-end server market. They marketed and and upkept the technology decently.
The acquiring company HP, already was in the server market with its own RISC (HP-PA/Apollo/68k) based platform without an identity. Remember, HP was now a major player in Itanium before it bought Compaq. Customers were promised an outlook in the future of continued development of the Alpha and Tru64.
Digital Unix/Tru64 and VMS are the fruits that ripen on the Alpha chip, never was it Windows NT.
Thinking about what happened to the Alpha makes me much happier that IBM did not buy Sun!!
Wow, I thought Itanium died when x86-64 launched.
(I do not work in the industry, so experts should correct me if I’m wrong.)
Itanium basically took over the space previously occupied by Alpha: little-endian, RISC-based CPU designed from the ground up for 64-bit mathematical operations and memory addressing. Windows and OpenVMS run on it just like they continue to run on Alpha.
It is a clean architecture which in *theory* can deliver exceptional performance on a variety of workloads, by having the compiler sequence and package up to six instructions per cycle. Unfortunately, the compilers never got good enough to fully exploit this instruction parallelization.
Now, Itanium is stuck in the same restricted market as Alpha — turn-key enterprise serving and scientific number crunching — while x86-64 takes over the general server market because it supports the legacy x86-32 architecture and makes no special demands of compilers.
Edited 2009-05-06 08:06 UTC
So they went from 100 to 118 units sold ?
Joke aside, the architecture seemed very promising , I recommend reading anandtechs article about the architecture.
The enterprise world however need backwards compatibility for its software stacks. Im not talking about cute flashy sites or clever company x, Itaniums are just to expensive for massive webclusters even if you do get mainstream software like win 2008.
And they dont have the pedigre like POWER or s/390 or sparc to make it which they belong the midrange to highend enterprise like Banks.
(im sure that they would absolutely kill any sparc performance wise, but thats besides the point)
Im almost impressed that they havent killed the arch yet.
You’d be surprised. There are only 9 IA-64 clusters in the Top500, none of which are in the top 10. There are two “pure” Opteron systems, one at #6 and the other at #10. The #1 cluster also uses Opterons.
Why? Because the price/performance and performance/watt of Itanium is awful.
The Alpha was in fact bi-endian, and the normal load/store instructions (the one which used less cycles) were the big endian one if memory serves (the cycle cost for little endian was small though)
I wouldn’t call an architecture designed for an ‘imaginary super-compilers’ clean..
According to the Wikipedia page on endianness, Alpha only runs in big-endian mode on Crays. However, I was not aware at all that Alpha had this capability — thanks for the knowledge.
As for Itanium, I guess clean != effective
Also, apparently it can do bi-endian as well …
Edited 2009-05-07 03:05 UTC
From what I understand Itanium has great FP performance but that’s it. It is only good for a limited amount of applications and has never performed as well as it was supposed to. I don’t think it has a very bright future to be honest.
Indeed, that’s what I’ve always heard. The fanboys need to elaborate more to convince us of Itanium’s goodness.