Dear Dale:
Re: "
>Alpha penetrated markets. Most of these were DEC shops running either VMS or Ultrix. Most of these sites switched to Alpha with little fanfare.<
Most certainly. Alpha was designed to replace the VAX, which was a Supermini, not a Mainframe. But, how many sites switched from IBM Mainframes to Alpha?"
Actually quite a few switched, mostly due to performance price ratios. And many more simply started using Alpha.
Re: "
>DEC's main advantage to IBM or any others mainframe systems was the ability to cluster. Alpha did this in spades. So your interpretation is quite flawed.<
This is only significant in that it enabled Alpha to continue what VAX started. For years there had been a paradigm shift occuring that gradually replaced mainframes with clustered minis. This did not kill off mainframes, but it did relegate mainframes to more and more specialized roles."
They replaced mainframes with clusters of minis. It was more reliable and in many instances even mainframes were clustered. You forget that many companies simply bought "Turn Key" systems. The VAR supplied all the hardware, all the software, installed and trained the users to run it. There are many more systems procured the latter way than systems custom built from scratch by the companies themselves. You seem to be of the old paradigm "DIY" rather than the one most commonly used "Buy it off the shelf". The latter method requires a lot of the former to be available. Very little exists for the Itanium. Much exists for x86 and the 64 bit RISC CPUs (BTW Itanium is not a RISC CPU).
Re: "
>And you forget that performance and compatability is a big draw. And price is a huge draw. Having all three could switch away all the OEMs from Itanium. There is no current market where Itanium Dominates. Some time in the future plans do not make much of a difference. OEMs are fickle. If Itanium doesn't outrun current systems, the OEMs could simply say that they will restart or continue their current RISC systems.<
So, Dell, Gateway, and SGI are simply going to continue with their current RISC systems if Itanium doesn't pan out?"
What??? Don't they sell Xeons now? Better tell Intel that they just lost 50% of their Xeon sales. Get real! IBM will simply push Power and RS/6000 systems, Compaq will sell Alphas (they still have two years left of production and could switch to Hammer just as easily by then) and Xeons, HP can just sell PA-RISC systems (they are still continueing development), Sun will still sell SPARCs and all of those Xeon and Athlon server vendors will still sell x86 systems.
Re: "
>HP has a new PA-RISC in the works, IBM a Power4 CPU, Sun has theirs and even Intel could go back to Xeon. And every day it looks like IA-64 is falling behind more and more.<
SUN is irrelevant because they are not going to adopt Itanium anyway. Well, I suppose that SUN could be relevant if Hammer is compelling enough versus Xeon that SUN decides to give the x86(-64) market another run.
While it is possible that HP will stay with PA-RISC and IBM will abandon Itanium altogether for their own Power architecture, this still does not address the needs/desires of the likes of Dell, Gateway, and SGI."
What??? Hammer will be available as an option. It does all that Xeon will do and adds 64 bit modes as a kicker. Also they could buy PA-RISCs, Power4s, PowerPCs or SPARCs instead. But they would be very mad at Intel for giving them yet another "dead end".
Re: "
>Just like many other "can't lose" propositions. Remember Microchannel?<
Who besides IBM thought that Microchannel was a can't lose proposition?
>How about Futurebus? Rambus?<
Only evangelists of these architectures believed that they were can't lose.
>Or how about the big x86 RISC replacement i860?<
Sorry, but when I worked at YARC I was one of the ones who recommended against going with this architecture. And, judging from the total lack of i860 support from mainstream vendors, I was not alone. If you were wanting to try to convince me that you had a point, you should have used the i432 as your example."
You didn't have blinders on then. It was your behind that would have been in a grinder, if it missed. You forget that Itanium has evangelists and you are one of them. Customers are not clamoring for Itaniums. It is being shoved down their throat and many are balking. That is one of the reasons the other developments are continuing. Intel thinks they can shove it down customer's throats. All it takes is one alternative and there is more than one to have people switch rather than take it. That is how Compaq got as big as it is.
Re: "
>There are many examples of something that supported by the biggest name in an area and although it looks like a "can't miss sure thing", it falls flat on its face because of irrational arrogance.<
But far fewer that have the top ten vendors in the industry behind them. Granted, SUN isn't behind Itanium, but SUN dropped out mainly due to political reasons."
Top ten vendors are not as solid as you seem to think. And if the customers don't buy them, then those top ten vendors will try to sell them something else. Or the customers will make another vendor be in the top ten. Let's see who has a backup plan still running. IBM has many competing lines (Power, RS6000 and Xeons), Dell still sells Xeons, HP sells PA-RISCs, Compaq sells Alphas, mainframes (forgot the Tandem line?) and Xeons, Sun sells SPARCs, Unisys sells mainframes and Xeons, SGI still sells MIPS and Cray still sells mainframes and supercomputers. Looks like there will be plenty to buy, if Itanium doesn't sell.
It doesn't have the whole hearted support from the top ten vendors of server class systems. It appears to have the support of "if the customers want to buy it, we can supply it." Thus, it is a customer driven environment. The competing lines got extended due to customer complaints and demands. They are still extending, so that shows how little demand exists from customers.
Re: "
>Take an example. A large company needs to replace its MRP/ERP system because the of the new goals of management. They do a search for MRP/ERP software that does all of the things they need. They find an ideal package with some extras that make it highly desirable. The systems it currently runs on are HP PA-RISC (5 yrs exp), IBM Power (3 yrs exp), Sun (4 years), Compaq (DEC) Alpha (6 yrs exp) and x86 SCO Unix (5 yrs exp). They are willing to write an Itanium version, but it will take an extra year and you would be the first. You know the drill, they will take the one with the highest experience, performance and lowest cost in that order. That gives them either Compaq, HP or a x86 OEM. Itanium would not be on the list (not ready yet and this company doesn't want to be a pioneer).<
Chances are that, if Xeon would be adequate for their needs, such a company will go with Hammer, if Hammer lives up to its promises. But, with the exception of the eight-way market that Intel is intentionally taking away from Xeon, Hammer will only be able to penetrate those market for which Xeon would be adequate."
Sorry to burst your bubble, but somehow you are hoping that having 64 bit extensions will not be used by the owners of such systems. Most of the problem is getting your foot in the door. "Hey, what will we use to back up the mainframe Oracle server? Hey, the Hammer system we have here can run a 64 bit Oracle server. Let's use that as it costs us far less and we know its reliable!" Xeon can't do 64 bit computing, but Hammer does not have that limitation. The number of applications needing 64 bit user code is small. Many just need the system services to be 64 bit like OS, memory, disk and RDBMS. The rest can be 32 bit with no trouble. A market of a million Hammers will get those things easy. Are you hoping that AMD won't sell a million Hammers? 10 million? That is a far larger user base than 10K Itaniums or even 100K Itaniums. BTW hammer could go into systems with more than 8 ways (an 8 way dual core Sledgehammer is 16 way core) by substituting each P2P arrow with a hammer with one HT link each direction pus a link going up or down and placing a plane above with same. Bingo, you have a 32 way (64 core) Hammer. Oops! Burst your bubble thinking that a Hammer with 3 HT links couldn't go more than 8 way, didn't I.
Re: "
> This result is why you cannot call IA-64 established.
Itanium is not established, it is entrenched. The difference can be illustrated by the military analogy of Bataan. MacArthur was never established in Bataan, but he was most certainly entrenched.
> And of the tens of thousands of applications out there (we are taking VARs, system integrators and software houses here) and even Intel with all of its resources could not afford to have the code ported to IA-64 (it would run into the 10's of billions easy and that is with a good optimizing stable compiler (and that is still not available) and reliable development environment).<
Major VARs actually prefer less common systems, such as Alpha and PA-RISC, not because they can do the job better or cheaper than more common x86 based systems, but because it is harder for their customers to switch to an alternative solution."
I have much experience in VARs and both potential market and what you already did have far greater impact than you evidently think. All have x86 versions because that is what the customers want. The others are because the x86 isn't big enough to handle it or the customer wants to standardize on a particular line (make it easier for them to support it). Take the warehousing vertical market. All three of the top VARs are based here in SE Wisconsin. Two of them use Oracle and one uses Informix (although they do use DB2, Sybase and RDB when demanded). Most of their non x86 customers use IBM (mainframe, power and RS6000), SUN, HP PA-RISC (9000 and 3000 series), and Compaq Alpha (VMS and TruUnix). None have an Itanium platform yet.
Re: "
>The closest historical parallel would be the 8080 to Z80 switch or the 286 to 386 switch. They did not require a recompile to run old systems faster. And they had a new mode that made them faster and gave them additional abilities that were very desired (the 286 to 386 switch vastly simplified the needed code models and made it easy to use much larger amounts of memory (no one I know would ever want to switch back to the 286 model and MMU)).<
Personally, I am hoping that the Itanium versus Hammer turns out to be the equivalent of NS32K versus 386, but I wouldn't count on it. NS32K had the 68K, Fairchild Clipper, MIPS, WE32K, and even NEC V series all as strong competition. If all that was needed was compatibility with the installed base, the 32-bit variant of the NEC V series would have far exceeded the 68K in market share."
Well how many existing CPUs can fight the Itanium. Power, RS6000 (PowerPC), Alpha, PA-RISC, SPARC and MIPS plus any of the specialty mainframe CPUs. Quite a few, even if you neglect Hammer. As to the V series having more market, did you forget that Apple, NCR, AT&T, Unisys as well as many others used it and its later variant, the 88K. Also you forgot that SPARC was available then as well.
Pete |
