Dvorak on Itanium

[Kosh Naranek]

Originally posted by Dogbert


Exactly what I thought. And Intel employee who knows "HIS FACTS".
When I'll see a McKinley review from Dell instead, I'll believe.
In house successes are woth nil.
You have a McKonley workstation which you didn't pay for, and which you propably don't even know it's maintenence cost.

I'll wait for a review from ZDnet anyhow. Besides, even if it's all but lies remember this: People don't easily beieve the creator. People easily believe "unbiased" sources. Even if they lie, it's enough that they repeat the lie that many times so people believe it's truth.

Intel has two problems, bad cpu and bad PR.
Too bad we don't have an AMD employee in the forums as well.

Wombat, think before you flame.

We don't need an AMD employee ... when we have you

Unbiased sources ..... NOT you .. that's for sure. ----> Do not blaim others for being biased when you yourself is.

Me being Biased .... ehhhh ... sure !!!
However, I'm not biased to a degree which is borderline fanatic !
I can can see good things on both sides which you obviously can't.

[Dogbert]

Sure I can see good things in Intel.
I currently use an Intel processor at home.

I just like AMD's business practice much better.

For me Intel is like Microsoft of the CPU world for the better and for the worse.

We don't want to start a Microsoft debate now, don't we...?

[isochar]

Originally posted by Dogbert


Intel has two problems, bad cpu and bad PR.
Too bad we don't have an AMD employee in the forums as well.

Wombat, think before you flame.

Care to back that up? The Itanium puts out fairly impressive numbers for a 733mhz processor with small amounts of cache, on a 266mhz FSB with DDR. McKinley gets the 400FSB of the P4, will probably clock in at over a gigahertz, and will have dual-channel DDR.

The Hammer is expected to be about 20% faster than the Athlon across the board, from all the posts I've read at Ace's. If you take that into account, a 3400+ Hammer will get crushed by an Itanium in SPEC. McKinley will be a whole 'nother story.

As for bad Public Relations from Intel?! What are you smoking? Let me have some.

Intel's main customers are corporations. It has huge relations with them, supports them at every turn, and give incentives. (Chipsets, anyone?) Even with end-consumers, Intel has good relations. Sponsors events, products are seen in advertisements, and its products have a good reputation.

What exactly is bad about Intel's PR?

I think you're confusing PR with COMPETITIVE PRACTICES. Which Intel's practices would be done by AMD if their roles were switched, so don't play the holier-than-thou BS.

[Dogbert]

Ok...


I see you did little reading about the Hammer series haven't you ?

For the price of 1 IA-64, you could build a 2-way hammer system and still have money left !

Besides, theh Alpha is currently a better investment price wise as well. Intel had a golden egg and they threw it away.


About the bad PR, after all the bad things everyone been telling about the IA-64 fiasco and after the info about Yamhill Intel still didn't do anything to get people interested in the IA-64 again.

IA-64 is nothing but a new fanshion. Exactly like the Cruso was.
BIG hype, and then...?

AMD kicked Intel's ass a long time ago with their 386DX40, ever since the K6 they've been back on Intel's heels.

Now it's P4 vs. Athlon XP. I really hope Intel have something up their sleeve to compete against the Hammer series, cause I like competition and constant price drops.

Till then, I really don't see why SUN and IBM should be scared of the IA-64. Too hot, too expensive, too backward incompatible... Too much of a fiasco.



Some food for tought:

Supercars are really great cars. Too bad they're way too expensive and most of them got bought by "mainstream" car makers.

The IA-64 is exactly like that, great CPU, yet not so practical.

[GNEP]

mmmm... 386DX40 - built a box with one of those in - ran like a dream, lasted for years and years and years (in fact, I think it's in the office of a small charity somewhere running windows 3.1 still!!). Give me a motherboard that can still run happily with a layer of dust millimeters thick on it these days!

[isochar]

As pointed out before, you need to take a look at this from the Itanium customer's point of view. (A corporation)

For the price of 1 IA-64, you could build a 2-way hammer system and still have money left !

Problem is that Itanium will run 64-bit code MUCH faster than Hammers. (SPEC will show that) McKinley will double up on those SPEC scores and compete with the best of them - the Power4 I believe. All while maintaining a substantially lower price point than all of its competitors, except AMD.

For the price of the 2 Hammers, corporations will be able to buy 1 Itanium. All while getting better performance, since any corporation that uses a server like this will have propietary software that is coded specificially for the processor.

Besides, theh Alpha is currently a better investment price wise as well. Intel had a golden egg and they threw it away.

Hardly. The Itanium was just to get Intel's foot in the door. The McKinley will offer top-notch performance in the server class environment, while doing so at a substantially lower cost than the competition. Even the Itanium's maintenance is 40-50% lower than comparable servers.

About the bad PR, after all the bad things everyone been telling about the IA-64 fiasco and after the info about Yamhill Intel still didn't do anything to get people interested in the IA-64 again.

How exactly is the IA-64 a fiasco? It didn't sell a lot of units, and that's about it. Again, corporations are the main customer. They DO NOT listen to rumors that Intel may be producing a DESKTOP x64 processor.

The point of the unit was to get Intel's server infrastructure in place. Once McKinley comes around, Intel will have already been serving out server chips for almost 2 years.

IA-64 is nothing but a new fanshion. Exactly like the Cruso was.
BIG hype, and then...?

Crusoe was a good technology, but they had little funding in comparison to Intel. Once Intel released the low power P3s it killed Crusoe's chances.

AMD kicked Intel's ass a long time ago with their 386DX40, ever since the K6 they've been back on Intel's heels.

This is the desktop market you're talking about.

Now it's P4 vs. Athlon XP. I really hope Intel have something up their sleeve to compete against the Hammer series, cause I like competition and constant price drops.

Till then, I really don't see why SUN and IBM should be scared of the IA-64. Too hot, too expensive, too backward incompatible... Too much of a fiasco.

Supercars are really great cars. Too bad they're way too expensive and most of them got bought by "mainstream" car makers.

This, yet again, is your problem. You need to look at it from a businesses point of view. Dropping a million dollars on a server is common place. The P4 is the mainstream processor.

[Dogbert]

Isochar,

Today is March 7, 2002.
Take a mental note for March 7, 2007 and we'll see by then the expensees vs. revenues for the IA-64, ok ?
I'll even get myself a hat for this special day...

[Dr Mordrid]

Lets look at it from the practical perspective for a bit.

The end result is where the rubber hits the road. One of the most demanding things you can do with a system is video editing, especially when it comes to rendering times and this is very much CPU dependent.

Recently a list was put up showing the relative rendering times of Premiere vs. MSPro when exporting DV and MPEG-2 to disk files, but this data also includes info that can be used to compare the AMD AthlonXP and the P4.

The MediaStudio Pro 6.5 info is most interesting since it is both SSE2 and 3DNow!+ enabled.

The result: the AthlonXP delivers rendering performance equal to or, more often, superior to the P4 at a much lower cost.

The same effect has been documented in 3D rendering software as well, both prosumer and professional.

If this real world performance gap in applications continues or widens with the 64 bit CPU's it won't matter what Intel or its webforum weenies say about their lab results or archetecture. They said the same things about the P4.

Dr. Mordrid

[Wombat]

but the server-class reliability just doesn't seem to be there.

Upon what basis do you claim that? I know you are on the McKinley team at HP, so it's hard for you to be objective, but what sort of servers are we talking about here?

I took a good look over AMD's hammer presentation. It looks nice for 1 & 2 CPU desktop systems, but the implications of having it in the 4- and 8-way setups are scary. If you lose a CPU, it appears you lose the memory it was controlling, and if you lose the CPU that is managing one of the peripheral buses, you're really hosed. Any server system has to be designed to lose CPUs gracefully, but at first glance AMD's design doesn't seem to stack up to current or future offerings from Sun, IBM, or HP. The big buzzword these days in servers is "RAS", and AMD doesn't seem to make it.
I admit that I'm also discounting AMD BIG TIME because I'm not aware of them doing anything more than 2 CPUs in a system before, while they're trying to break into a market where everyone else has decades of experience in major multiple-processor systems.

The problem with this line of reasoning in the first place is one of cost for performance. To implement an Itanium system, one must purchase a very expensive and proprietary (for now) set of hardware components. Then there's software. It all has to be brand new.

To get any solution in the door requires proprietary hardware. That's a given. System pricing is also something I'm unfortunately gagged about, but I will say that I'm not worried about it.

Any new or proprietary software must be rewritten for IA-64 to take advantage of ANY of its true benefits (the many registers and massive parallelism), and this makes it a very difficult architecture to compile for. I would even say that it may be prohibitively impractical to realize Itanium's performance potential, due to its architecture's incredible complexity.

Software would have to be adapted to get the most out of whatever system it's going to run on.
Regardless of architecture, if a program is written with 32-bit data widths in mind, it's going to have to be tweaked to get the most out of a system with a 128-bit bus.
For the programmer, IA-64 is actually a fairly simple ISA. I've found the assembly code to be extremely readable, I really don't think it's complex. Lots of little, easy-to-understand pieces. Besides, at the application level it's no different than supporting Sparc and Power architectures, but companies gladly do that. I also don't understand what you mean by "difficult to compile for." Compilers are provided, by Intel. You've noticed how Intel has released some nice x86 compilers lately, right? That's just runoff from what has been bought/learned in order to make good IA-64 compilers.

There are massive costs associated with migrating to the IA-64 platform. With a Sledgehammer system, you would have a much lower hardware cost to implement the system, and the entire legacy of X86 to draw upon for software, including the many varieties of X86 Unix and their associated applications. Even if we go with your assumption that the Hammer system would be less reliable, the total costs to the customer would probably be less if he purchased TWO Hammer based systems running redundantly than a single Itanium system.

I don't think Hammer will win on TCO, but we'll see what they do to keep maintenance cost low. To server customers, their data is far more valuable than their hardware, and downtime is death. Nothing in the Hammer presentation even hints at things like chipkill, redundancy, rerouting, throttling, self-maintenance or self-awareness. It's not something they've done before. Also, how could you leverage the "x86 legacy"? That stuff is all going to be 16 or 32 bit, designed for single-processor systems, and would have to be rewritten anyway to address the same concerns you have about programmers handling IA-64.

IA-64 is not for your average consumer. It's not even for your "cutting-edge" consumer. It's for the same guys that have been buying the big boxes for years. Anybody that thinks that they will have an IA-64 as their home desktop over the next few years has made a grevious(sp?) mistake.

The amount of time and resources Intel has poured into IA-64 over the past 10 years boggles the mind.

Agreed. But Sun, IBM, and HP always do the same kind of thing to launch a new architecture. A new architecture from white paper to silicon is 5-7 years. These big commitments are not unusual.

If this whole architecture was never destined for anything but the very high end of the server market, Intel has wasted their investment. They will never sell enough volume in this super high end market to justify the billions spent.

Never said that. No comment. I'm just saying that IA-64 is <I>capable</I> of playing in that club, while I don't see anything that leads me to believe that AMD can cope.

In my humble (but realistic) opinion, Intel management has simply been pig-headed about the whole affair. This is their baby and they don't want it to die no matter how ugly it is. And to think that they just bought the most impressive chip architecture I have ever seen in the Alpha EV8 design and are going to just kill it.. the foolishness of this boggles the mind.

It's HP's baby too. IA-64 has its roots in the HP WideWord architecture. EV8 was cool, but a lot of other stuff out there, and EV8 wasn't going to be anything earth-shattering. Besides, the Alpha engineers are working on stuff just as cool, I'm sure. Even after being here in the lab a while, I still find stuff that makes me say "holy shit, we can do that?" It's really cool to implement things that "expert analysts" say are impossible.

(Sorry if this comparison offends you, Rob.. but it's really not your ugly baby.. you just have to babysit for the thing.)

Yes, actually, it is my baby. Intel puts its label on everything, but there's a lot more of HP in this than most people see. It's a shame that Merced was the first thing people saw, but we couldn't do anything about that.

Intel clearly meant for IA-64 to take over the desktop around the 2010 timeframe. If not, why even mess with it? I would love to have something better take over for the X86 architecture, but I really don't see it happening anytime soon. I had been rooting (in my wildest dreams) for Alpha to take this role.. what a shame.

I can't comment on future roadmaps. I have the utmost respect for my peers in the Alpha group, but I never ever expected there would be an alpha in home PCs.

In comparison, people that buy servers buy whatever meets their needs the best, and the software is generally custom-tailored to that.

Again this favors Hammer over Itanium. On both counts.

I really don't see how you figure on this. People hop from Sun to IBM to HP or whatever whenever one meets their needs better. Data is generally architecture agnostic, and if I want to switch from Oracle running on Sparc to Oracle running on PA, it's not painful.

One of the very moot points people often bring up about Hammer is that there are no operating systems for it. Well it runs anything 32 bit, including Windows, faster than any other processor, plus check this out.. Yes, boys and girls, there is an AMD64 Windows XP. Linux and FreeBSD will be available too, in 64 bit versions for Hammer. And programmers have said how wonderfully easy X86-64 is to program for..

Software wise, IA-64 is in better hands right now. Windows, Linux, HP-UX, and other commercial flavors, all run on IA-64 right now. You can even go get your <A HREF="http://www.matrox.com/mga/support/drivers/files/ia64_contact_sales.cfm">Matrox drivers</A> for IA-64. Last time I checked, there aren't any more commercial Unices for x86, since SCO is getting out of the business.

quote: Sorry, NDA.

You mean everyone else is allowed to talk about Yamhill except you? You don't even work for Intel, Rob.

There's a lot of shared IP. I don't know anything about Yamhill. I just meant that I can't talk about future IA-64 markets.


Oh, and sorry I didn't address much of your post Doc, but I just want to say that you can extrapolate pretty much nothing about IA-64 based on what you see with P4s, or anything else that you might have on your desktop. And I certainly don't appreciate the implication of being a "weenie," nor have I ever worked for Intel. I laugh at the P4 at least as hard as the rest of us.

[KvHagedorn]

Thanks for the reply, Rob. You know, I was all excited about IA-64 when I first heard about it, but then there were all the delays and when Merced came out so many problems became apparent. I guess we will have to wait and see. I'm glad there is someone out there trying to compete with Intel now that they have engulfed Alpha, and I hope Hammer can do it. I see your point about the memory controller being on die. It might make Hammer faster, but there are drawbacks too. One other one that worries me is this. What happens when DDRII sets its standards officially and comes out, being non-backward-compatible with DDR? I hope AMD has allowed for this with the increased pin count on the Hammers. All in all, Hammer will probably turn out to be a worldbeater in the workstation arena, but you're right (and I agree with you here) that to implement a credible server platform, a company needs to dedicate itself to much more than CPU design. It will be interesting to see how AMD handles this step. I have been very perturbed with their reluctance to make chipset production a priority, though they do make their own dual chipsets.