Intel’s high-end Itanium chip is not making the kinds of inroads that Intel had hoped. It’s Xeon chip still rules the roost for servers. In response, the company will be launching a program that lets potential users try out Itanium hardware before they buy. A ZDNet article covers where Intel’s going with the Itanium and Xeon.
Linux on POWER. I’d love to get my hands on some IBM hardware for desktop Linux.
My Athlon 1GHz processor has been good but it seems that the little guys are forced to use X86. I like how IBM has applied the FOSS model to hardware, not just software. Instead of clones, we might actually see some unique hardware that is designed to support renegade systems.
With the x86 but like MS they are having a problem executing outside of their monopoly (which i define as x86 processors).
They got luck and now they keep screwing up.
Look, I’m a theoretical chemist, and we all know that the Itanium is probably the best chip out there for calculations right now. The problems though are legion. First, it costs nine arms and ten legs. Second, the chips can heat a ten-story building.
However, those can be overcome. I, in fact, had a chance to try out one of those HP workstations about a year ago. Frankly, that sucker has kept me or anyone I know from getting one. The main problem for me was due to Intel not having a good linker for Linux. I mean, efc worked fine (didn’t optimize great yet, but worked), but used ld for linking. And, unfortunately, it had the side effect of not allowing many programs to link. Something to do with too many COMMON blocks or the like. Then you add on the horrific cost (only matched by the EV7), and amazing power/heat…
And now the problems are getting worse. The EM64T, or whatever Nocona is, is showing amazingly good floating-point. Way better than current workstation champ Opteron. With that, I’m not sure anyone will look at the Itanium. There’s a big split between Opt/Noc and then POWER5 which is the big iron most use.
Fortunately for Intel, the massive investment in Itanic, which is basically a 100% wash, has not impacted their share price at all.
Its simple – Intel built a new architecture and no one showed up. I’m not saying Itanic is not fast or useful, simply that it has been a market failure. The adoption rate for this architecture is basically flat.
Sales of Itanium systems will be dramatically boosted when Intel releases a common architecture/socket for Xeon and Itanium processors, and the Itanium then becomes a feasible upgrade option for those who already own Xeon systems.
I recently read in iSeries magazine that the POWER6 will be taking the approach of not going a multi-core route so software companies won’t kill you with licensing fees by charging for each core. However, the processor (which i believe said would be produced with 65nm tech) will be a square foot in size and be equivalent to about 16 POWER5 processors. In other news, I may actually be able to buy a quadcore xeon.
POWER5 2702
Itanium 2 2161
Everybody knows that the Itanium has a better overall architecture than the x86’s, but it can be tamed with an higher complexity of the chip, which will be compensated by the huge production numbers and AMD’s competition. The PowerPC and Alpha’s superiority of the ’90 didn’t helped to get rid of x86s. The difference is that an Intel product is killed by another intel product. ( It occured before with the i432 and i860 I presume )
The fact is that Itanium is already obsolete as it won’t keep its performance edge against cheap x86s. Intel needs to create a more advanced architecture.
( A 6MB cache ! It makes quite a difference, what would be the performance of a 6MB cache Xeon ? )
A common socket might help, but I don’t think the kind of people who upgrade CPUs would buy an Itanium. Home powerusers upgrade CPUs for the latest games and whatnot, but how often do servers get upgraded besides RAM and hard drives? Swapping CPUs is risky, even more so when you’re swapping instruction sets. I imagine most places would buy a new Itanium server, test it, then if it works swap the whole thing. A somewhat recent slashdot article made the same point.
I think Intel launch a program where their chips are cost-competitive with AMD’s chips.
“A common socket might help, but I don’t think the kind of people who upgrade CPUs would buy an Itanium.”
I work in scientific computing, as does The Matt (hello from CSU!). It would certainly be worth it to us to have an upgrade option down the road for any system we purchase, and purchasing a large cluster of Xeons and down the road upgrading it to Itaniums is a very tantalizing prospect for us, as we could upgrade our existing infrastructure rather than throwing the baby out with the bathwater. We have the same upgradability concerns with dual core processors.
The fact is that Itanium is already obsolete as it won’t keep its performance edge against cheap x86s. Intel needs to create a more advanced architecture.
Thats a common misconception. The Itanium Platform was developed exactly for that reason: to have a new architecture that can still be optimised in the next 10-15 years. It gets harder and harder for Intel to find better chip designs for IA32. Performance improvements are merely accomplished by shrinking the die size and increasing the clock frequency.
I agree that the new 64bit extensions for x86 have changed these facts a little bit; the new (more) registers and new instruction set definitely allow room for some new optimisations, however I severely doubt that the design is as forward-looking as the one of the Itanium.
However as good as the Itanium might be, it might not have a bright future if Intel/HP don’t manage to make some serious profit with it.
Intel is a great semiconductor company. They have some of the best fabrication plants in the world, and more capacity than all but a handful. But that has always allowed them to get by with flawed designs. The aerospace equivalent is that anything can fly with enough thrust. Intel is good at thrust, and they’ve made pigs fly.
The success of the x86 was a fluke. Without the IBM PC, it would have been a modest success as an embedded processor. Intel’s first 32 bit processor was a complete flop. Their first RISC processor was a flop. Their most popular non-x86 processor is derived from the StrongARM that they picked when they bought Digital Semiconductor. Their first VLIW processor was a flop. Now they are on their second generation, and it’s getting stiff competition from an AMD x86 design that does what Intel could have and should have done years ago.
Very Long Instruction Word is an interesting concept. The idea is that you move optimization to the compiler, counting on it to arrange the instruction order so that the processor’s execution units remain busy. But the compilers don’t seem to be delivering on that promise. Instead, the focus of compiler technology has shifted to JIT compilers and run time optimization. Maybe someone will figure out how to make VLIW work. It may take a new programming method and language. My guess is that the first strong VLIW design won’t be from Intel.
I remember back in ’97 I bought a 300 mhz PII (which I believe was top of the line x86 back then) and thought to myself my next system will be a Itanium…or whatever the code name was back then.
Another problem has/had is that the compiler had to do all the instruction re-scheduling to keep that pipeline full. Supposedly, this was/is a bitch of a problem.
I work in scientific computing too. My adviser got an Itanium 2 system last year and it’s been nothing but a waste of money. It’s not that the hardware isn’t good, it is. It’s just that the compilers for it are terrible because they’re apparently so hard to write. Basically, my adviser’s code sometimes runs (really fast), sometimes crashes randomly, and sometimes doesn’t even compile because optimizing his code (~40,000 lines) requires more than 8 gigs of ram! So he can only run it for the cases that don’t make it crash, which are (unfortunately) the parts that don’t require 64-bit ram addressing (totally invalidating the main reason he got it). To make things worse, he has the Intel premier support for their Fortran 90 compiler and it’s completely worthless. He gets about 1 message a month where some guy in Asia they outsourced support to says, “I just got back from vacation and I don’t understand what problem you’re having.” The worst part is that most scientific guys use Linux, which is still flaky on Itaniums — his machine randomly failed and he had to completely reformat. Bottom line — just get a couple of Opeterons or PowerMacs instead and run clustering software.
It’s just that the compilers for it are terrible because they’re apparently so hard to write.
Intel compiler is not so bad. If you are looking for an opensource compiler, wait few feeks for openImpact. A beta version will be released. Keep an eye on http://www.gelato.uiuc.edu/
The worst part is that most scientific guys use Linux, which is still flaky on Itaniums
If you need help on Linux over Itanium, just send post to gelato (http://www.gelato.org/). You’ll find people not in vacation who knows a lot about Linux on Itanium. You’ll find here kernel developers and scientifics.
Goodbye itanic hello amd64.
It may take a new programming method and language. My guess is that the first strong VLIW design won’t be from Intel.
Well, part of the problem is that there is only so much concurrency you can wring out of sequential C code. The Itanium2 is massively parallel: 6 integer units, 2 load units, 2 store units, 4 MMX units, and 4 floating-point units. It is nearly impossible to for C compilers to keep this many units full, except in very special cases (heavily-numeric code).
One way to extract more performance out of highly-parallel and multicore processors would be to move to functional or concurrent languages. Functional languages minimize or eliminate side-effects, which allows the compiler a lot more flexibility in parallizing execution, at the cost of more data copies. Concurrent languages explicitly encode parallel processing within the language, eliminating artificial dependencies between program threads.
Sales of Itanium systems will be dramatically boosted when Intel releases a common architecture/socket for Xeon and Itanium processors, and the Itanium then becomes a feasible upgrade option for those who already own Xeon systems.
That may be helpful for a small number of people but I don’t see how upgrading from x86 to Itanium is helpful for most people. You would need to use entirely different code to run on Itanium. What good is it if you have to rewrite your entire codebase? Sure you can run it emulated but why would you if you already have a working processor that costs much less and performs just the same.
I don’t get it. Why are you spending many many thousands of dollars on hardware and not getting a fully supported, company backed OS that is HPUX? If you are not buying HP Itanium boxes, whom are you buying from?
Sales of Itanium systems will be dramatically boosted when Intel releases a common architecture/socket for Xeon and Itanium processors, and the Itanium then becomes a feasible upgrade option for those who already own Xeon systems.
x86 and IA64 (Itanium) are two completely different architectures. You can’t just plop an Itanium in a Xeon board, boot up and keep on chugging, even if they have the same number and arrangement of pins. It’s almost like trying to put an Athlon XP in a PowerMac G5 board.
Water flows downhill, sun rises in the east…..
Itanic hasn’t met expectations for a long time now. It stems from Intel’s steadfast pattern of releasing new chips and calling them “server” chips. If they’d been committed to eventually switching everything to the IA64 instruction set, they would have really hammered Microsoft until MS got Windows right for it, and dropped most, if not all, PC x86 production.
They could have made it happen years ago, but they continued selling Xeons. Now they don’t have a choice but to go along with x86-64.
And that’s bad for everyone, really. x86 needs to die like m68k did at Apple.
m68k wasn’t all that bad.
x86 and IA64 (Itanium) are two completely different architectures. You can’t just plop an Itanium in a Xeon board, boot up and keep on chugging, even if they have the same number and arrangement of pins. It’s almost like trying to put an Athlon XP in a PowerMac G5 board.
Actually I2 emulates x86, but the price is so prohibitive there really is no reason to switch to a processor that costs n times as much as an x86 just so you can emulate x86.
http://www.intel.com/products/server/processors/server/itanium2/ind…
x86 and IA64 (Itanium) are two completely different architectures. You can’t just plop an Itanium in a Xeon board, boot up and keep on chugging, even if they have the same number and arrangement of pins. It’s almost like trying to put an Athlon XP in a PowerMac G5 board.
*cough*
Google: itanium xeon common socket
Time: 2 seconds
Result: An end to cluelessness!
http://www.eweek.com/article2/0,1759,1565228,00.asp
Well, a while ago the Athlons and Alphas were supposed to end up having a common socket architecture. At least the Athlon ended up implementing the Alpha’s bus. The idea was to allow manufacturers to offer motherboards that could be used with Athlons or Alphas depending on the depth of the customer’s pocket and computing requirements.
….we all know how that ended up.
Even if you can just pop an Itanium in a commodity motherboard there are a few hurdles to overcome:
1. Possible incompatibilities due to the differing ISA’s.
2. Added cost to the motherboard.
If you want to run x86 code, clearly #1 hurdle makes you think twice since what is the point of buying a processor so significantly more expensive that runs legacy code at a fraction of a much cheaper part?
And if you are running a Xeon machine, chances are that is because you are not willing to go and pay for a more expensive Itanium, so the added cost from #2 is not a good thing. Even if you could actually upgrade it is not worth it, since the motherboard will be obsoleted in just 1 year. So you must as well stick to a xeon only part.
POWER5 2702
Itanium 2 2161
Thats what you get when you run a single CPU benchmark on an
E class system and give one core exclusive access to a 36MB
lvl3 cache.
*cough*
Google: itanium xeon common socket
Time: 2 seconds
Result: An end to cluelessness!
This is OSNews. Cluelessness is a way of life here, and the natives get very grumpy if you try to fix that.
> Sales of Itanium systems will be dramatically boosted when Intel releases a common architecture/socket for Xeon and Itanium processors, and the Itanium then becomes a feasible upgrade option for those who already own Xeon systems.
Wishful thinking, what’s the point of buying Itanic gear when you can buy similarly capable but far more proven peice of equipment from Sun (UltraSparc) for half the price? Plus as a bonus with Sparc you get to keep full binary compatibility for any Solaris/Sparc software that ever existed. Sorry, but the Itanic camp is in BIG trouble. There are only two real contenders in the 64-bit space — SPARC (Sun/Fujitsu) and Power (IBM) — and this picture won’t change much with an exception of Opteron taking Intel’s place.