Slashdot Mirror
Intel's Itanium 2: Succeed or Fail?
An anonymous reader writes "'Intel's most powerful processor ever has the ability to take on IBM, sink Sun, make or break HP, and crush or revive AMD,' says Fortune's David Kirkpatrick. But the 64-bit question is what happens to the heavyweight competition if Itanium 2 succeeds or fails?"
As Intel now loses its backwards compatibility, they also lose their biggest advantage. Sadly, the IA64 will probably lose out to less spectacular, but IA32 compatible designs.
Alpha tried to emulate the x86 earlier and failed. Sadly.
Itanium/2 is a 64 bit processor. So it needs 64 bit software, including the OS.
Umm, no. For example I am running 32-bit Solaris on a 64-bit UltraSPARC. And applications compiled 32-bit.
Whereas in the case of Windoze, the 32 bit stuff (and even some 16 bit stuff) is built right in to the API.
Yes, that's why it's called the Win32 API. Work is well under way on Win64, but in Microsoft's ideal world, almost no-one will write to the Win64 API - they'll target the CLR, which itself will be 64-bit native.
Then the millions of apps that people use, right now an excellent way to lock customers in, are going to turn into a lodestone around their necks.
Yes, just like when Apple moved from 68k to PPC? Nope, wasn't a problem.
I'm sure Micro$oft is pissed as hell, but Linux is going to take a huge upswing when Itaniums start flying off the shelves.
That doesn't necessarily follow either. After all, Win 3.11 didn't fully exploit the 80386 either, and it wasn't 'til the first NT that Microsoft did.
It's already happening, you just haven't noticed it yet.
Jon.
I think the author has it backwards. He says if the new chip fails then blah. However, the more coherent argument is that if blah happens, then Intel's new chip has failed.
But, the author doesn't seem to realize that there's more than just out and out success or failure on the spectrum. It's more likely that there will be incremental change. Intel sells X units to A, B, and C, AMD sells Y to D, E, and F, and IBM, SUN, and co. sell to whomever. And things kinda ballance out.
All this new technology that's supposed to change everything dramatically, changes things to the degree that it's touted to. My money is still on evolution rather than revolution.
In mathematics, one does not understand things, one merely gets used to them.
--VonNeumann
Plus, less power consumption could mean thousands or tens of thousands depending how many servers you have. If you're google or some other huge site with thousands of systems, the power savings means lower operation overhead.
Well, the trend broke in 2001 when people started to notice that the machines needed for this generation of software was not the fastest but the slowest machines on the market. That most users did not need a top end machine and instead could buy the slowest processor out there. During 2002, the same came true for lap tops. Now everyone is swimming in so much wasted CPU power that it is going to finally crush those that can't adapt to radically lower needs compared to what Intel and their competitors are churning out. Ask someone who runs a computer room and they will tell you that they are quickly consolidating old servers that cost $250K three years ago to a server that costs $15K and only takes up a quarter of the room.
Intel is in real danger of not surviving because it does not understand where we will be in 5 years. 5 years ago when they were in the middle of this effort they did not see our need for speed slowing dramticly and are now producing a chip that has such a limited market that it will never be profitable with all the investment that was in put in.
When you look at how a company responds to the typical S curve of development, they may make the first curve but often that screws up their timing on the second curve and they just go off the cliff.
It is something of a question whether this change will open up opportunities for new software. I think it will. Think shared memory -- very large memory spaces being simultaneously updated and accessed by multiple independent processes and processors performing different tasks possibly for different users. The three drivers of technology are corporate databases, games, and pornography. Huge memory spaces with multiple processors attached have many possible breathtaking applications in each of these domains. Start coding.
I think the success of the Itanium not only rests on its technical merits but more importantly it rests with how much mindshare they can get for the product with the business people who, more often than not, end up making technical decision in a void.
I think that Intel is aware of this. Marketing can make the product. The best engineered solution does not always win out.
...Sigh here we go:
First, it's too hot and too expensive.
The people who buy these things know this and can deal with it. Remember, these are not crammed in like Mini-ATX towers (like the one under your desk). They're deployed by professionals in a professional environment with standards for this stuff.
Secondly is doesn't have any applications. I don't mean Gnome and KDE, I mean the sort of applications that big corporations run.
Big Corporations can and will port their existing (probably already 64-bit) applications to Itanium to take advantage of the newer / faster platform. ISVs are already porting applications to it and have been for a while.
Thirdly it isn't backwards-compatible with any existing architectures. You can't just take your binaries over and run them, at least not at full speed. Applications will need to be ported and retested. This is not insignificant in time, effort and cost.
See above. Porting will and has happened. If the logic can be presented that the company will either save or gain money by upgrading to this hardware then it will happen. It just makes business sense.
Fourthly, most people who want 64-bit in the corporate world already have it in the form of SPARC, Power, PA RISC and Alpha. Why should they change to an unproven, immature "jam tomorrow" architecture given their working and reliable systems already in use?
When the systems already in use are cost prohibitive to maintain they will be abandoned. A smart company will see the trend and start migration early. The Sparc platform is dated and loosing it's performance edge very quickly. The IBM Power series is still a reasonable choice. PA RISC who? Alpha who? You need to understand that IT departments invest for the long-haul, you won't see too many more shiney new Alphas being purchased not because they're bad but because C[T|I]Os know they're a doomed platform.
I'm afraid intel missed the boat by about 10 years. If they'd brought out a 64-bit RISC at the same time as SPARC, MIPS, Alpha and Power they might have stood a chance.
Or they could be going under like so many of the platforms you just mentioned. The 64bit world is certainly not new but it definitly requires some re-thinking in todays world. Intel is in a great position to do that.
I don't even need to mention how Athlon 64/Opteron will eat its lunch in the commodity sector of the market.
You don't need to say it because you can't say it. At least not yet. I too doubt that Itanium will be a hugh smash in the commodity arena. Not because it's inferior (I'm not arguing that either way) but because the money isn't there.
The companies that need and use 64-bit applications will not want those applications running on commodity hardware. They'll want a well supported platform and one that works time and again. Itanium can provide this. IBM can provide this. AMD cannot - they don't even make their own motherboards for christ sake.
Frankly, once a company has enough business to justify a 64bit platform they'll probably be profitable enough to deplay a good one - not the one from CompUSA.
Sun are still scheduled to release the UltraSPARC-IV this year (at last report) which will be dual-core (same as Power4) and might again leapfrog IBM for a while.
Sun aren't doing that badly, all things considered, given the current state of the economy. We'll see how things pan out over the next few years, but it's too early to say Sun/SPARC is dying.
The main problem is that the theory sounds good, but the reality shows a LOT of problems. The Itanium is a VLIW processor, which is quite different from the more traditional RISC and CISC designs of other chips. The idea is rather similar to the difference between RISC and CISC (which, these days, are more or less the same thing), move more of the optimizations into the compiler to make the chip design more simple, thereby allowing more money to be spent on fatter pipes, bigger caches, etc. for the chip.
The problem though, is that it's often EXTREMELY difficult for compilers to effectively optimize software for VLIW chips. Since the Itanium has no out-of-order execution or branch prediction, these things have to be done entirely at compile time. The compiler needs to organize the software so that the chip is constantly being fed with data rather than having the chip dynamically rearrange some instructions if others are sitting waiting for data. It also needs to include it's own concept of branch prediction, suggesting which branch is more likely to occur. What's even worse (and which I rarely see mentioned) is that it has to optimize it's software for a particular chip design rather than an architecture, ie Itanium software needs to be recompiled for the Itanium2 in order to see many of the benefits of the new chip.
As far as manufacturing goes, that's comparatively easy for Intel at least. There they just have to put up with a huge die and extremely high power consumption. Not exactly a cheap chip to manufacturer, but manufacturing chips has always been Intel's specialty. Also, the high selling price of the Itanium means that Intel can afford quite a bit of leeway.
Anyway, long story short, the big problem with the Itanium/IA64 in general is that it's a design that is VERY difficult to optimize code for. It requires a very good compiler to begin with, but even then there are simpily some optimizations that just can't be done at compile time, and those situations will hurt the performance of the IA64 chips a lot. If Spec CPU2000 scores are anything to go by, the things from CINT (ie databases, compression, FPGA design, compilers, etc.) are much harder to optimize for IA64 than CFP (mostly scientific computing applications).
Don't be so quick to predict the demise of Itanium. I would question your analysis based on past history of Intel products.
1. Heat - has been an issue since the 8087 and lower power products or improvements in heat removal technology have continuously become available. Even in current Itanium/Itanium 2 (Itanium Processor Family - IPF) products, heat is an issue but not one that is preventing IPF products from shipping. Over time you will see a significant reduction in dissipation in Deerfield/Monticito (SP?) but, in any case, solutions to the heat issue are becoming available.
2. Cost - Intel products are only expensive while customers are willing to pay high prices for them. Any time Intel has had competitive pressures, they have been able to drop the price to meet the new price point OR introduce new products that allow them to maintain their margins.
3. Nobody seems to understand that there is an IA-32 processor core built into the chips (starting with McKinley (Itanium 2)). For backwards compatibility, it's really an operating system issue more than a hardware/software emulator issue. When the operating systems are properly implemented, IPF will be able to run 32-bit IA-32 applications concurrently with 64-bit IPF applications. When Linux supports this, I think you'll see interest in Hammer wane.
4. I would disagree with your comments on the people who want 64-bit already have them. I would not disagree that there are limited projects testing out different 64-bit architectures, but I would be very surprised at there being any large server farms out there with the latest incantations of Power or Alpha and the SPARC/MIPS are probably looking for an upgrade.
5. Itanium is ideally suited for Linux. I agree with your comments with regards to Windows - but when you are upgrading to a new Linux release don't you rebuild/retest the application to make sure it still runs? In our Linux systems we have been able to port directly from IA-32 to IPF without any changes to application software.
I believe that there is a lot of opportunity in the market for a "standard" 64-bit processor and this is what IPF is designed for. IPF may not be the best or the first but they do have the track record in taking over a market and maintaining it. Nobody has made a lot of money betting against Intel and nobody has ever gotten fired for choosing their products.
I still think that's true. Windows on Itanium is a terrible value proposition -- almost nothing will be native for years and years to come, and x86 execution mode is way way too slow to be cost effective. I think we'll see very little Windows on Itanium.
OTOH, Itanium is virtually ideal for vendors moving from proprietary chips/UNIXen to Linux. I was still fairly skeptical about Linux' chances back then, but I'm not anymore. Linux on Itanium is going to be a smash hit and will dominate the datacenter.
Windows on servers is ... iffy. I see the possibility that AMD's x86-64 will be a hit in that market, but you'd have thought Athlon would be interesting too and it was completely ignored. Then again it's Microsoft's only real chance in the large server market so you can count on them pushing it really hard. If they succeed then expect an Itanium with a much improved x86 execution mode; I don't think Intel will go the extended-x86 route. If AMD does not succeed then Windows is going to be pigeonholed as a small server.
Regarding other chips, only POWER looks set to survive/thrive, but only in traditional IBM environments. Sun is in the middle of a financial collapse; I would be surprised if we see more than one additional generation of SPARC technology from them. Fujitsu has a nice SPARC, years ahead of Sun, but SPARC stuff is such a bad value proposition these days that it and Sun are going to die fast.
jim frost
jimf@frostbytes.com
Not all apps can be run on clusters of cheapo Intel! Who's *actually* running Oracle RAC on Lintel, really? Who, apart from Google, runs a set up like Google?
For example, if you're a Sun customer, like and use Solaris, you buy the low cost Sun boxes for the front end and rely on warranty for your maintenance needs. When you have an app that benefits from larger SMP boxes, you buy the larger SMP boxes and maybe rely on warranty if they're horizontally scalable apps, like say, app servers, running on 480s or v880s for example.
On your really large SMP apps, campus clusters, massive server consolidations, mainframe replacement F15Ks and the like, you buy a service contract.
Cheap Intel hardware isn't some amazing panacea that's going to replace large machines. Theres's also the cost of managing all those small boxes to take into account. A couple of reliable and available SMP Unix boxes can well be more cost effective than lots of little boxes.
I don't see how a new processor from Intel, or anybody else for that matter, is going to cause "serious competition" for any vendor such as HP, IBM, or Sun. When choosing a solution, IT doesn't go for Sun because its run on a Sparc CPU. They don't choose IBM because it runs a PowerPC. I give up on why they choose HP. :)
The point is, the CPU is just 1 little part in a solution. Intel isn't going to do any damage to these vendors unless they supply the entire solution, which isn't their business! To think otherwise is pretty dumb and a bunch of PR bullshit attempting to inflate Intel's stock value.
A slip of the foot you may soon recover, but a slip of the tongue you may never get over. -Benjamin Franklin