[Poll] Opteron / Hammer => "Too much to cover for it to work?"

[shathal]

With the "advent of Hammer" (Well - ETA of ca. 4-6 months before any real product can be actually gotten ahold of), I was wondering what other people would think of my opinions on Hammer. Hence - this post & thread .

My general point of view is a little concerned one, as I generally dislike it's Marketing Hype (but then, I dislike marketing hype per se, so it's not much of a surprise). I will go into detail as to why I am more than a little concerned that Hammer may not be the "saviour" of AMD that it's made out to be.

Hammer on Desktop => 64 bits "for free"?
==============================
Well - you'd like to think so, wouldn't you? However, MS has as yet not really admitted to dedicating 64-bit support to AMD's new chip (they've been rather quiet, though I remember reading that AMD seem confident). However - the problem is not only the OS. You still need drivers. Lots of them. EVERY bit of your hardware needs a new 64-bit driver for this platform.

The reason why I am a little concerned is that with a market share of less than 20%, trying to get the hardware vendors to write drivers for 64-bit OS that may not actually come about until MS decide that they DO want to support Hammer, is a bit much. Furthremore, getting hardware vendors to do ANYTHING (like accepting/using new microcode instructions such as 3D Now! and SSE/SSE2) is quite a feat. Intel has considerably deeper pockets than AMD & thus has got a lot more dollar-muscle to throw at the industry to get things accepted. AMD just can't parallel that.

64 bits on Desktops => "Who needs it anyway?"
==================================
"If I get 64-bits, that means I'm twice as well off as with 32-bits, right"?

Wrong. 99% of users/apps just don't NEED 64-bit stuff right now. You can quite happily go up to 64 GB of memory based on 32-bit stuff (with a certain bit whose name I forgot just now) - and unless you actually use 64 bit APPS, you're none the better off. And this means that the Software vendors write 64-bit apps FOR a MS/AMD OS. Pardon my cynicism, but rewriting software "just" for someone who has less than 20% market share (and it doesn't look like it's getting bigger anytime soon) seems a wee bit risky.

Also, there's the problem of validation (I'll get to that later) of the whole platform that may be a problem, but more on that later.

At the moment, SOME (!) Servers need 64-bit power, that is true. 64-bit for the Desktop is simply pointless at this point in time (and for a few years to come, IMHO). Even most servers should be quite fine with 32-bits. I do not mean to make this sound as "Yeah, 64-bits is a stupid idea - let's stick with 32 bits!". No - that's not it. I merely try to highlight the fact that 64-bits (and all it entails - hardware/software-vise) just is not worth it at the moment for a good 99% of Desktops/Servers out there.


So - Hammer gets to play with the server big boys? Finally AMD can slice into the Interl-server market, right?
=============================================================================
Well - you'd like to think so, wouldn't you? The problem is that - if you know anything about servers - is that Intel is not by ANY measure the only player in the arena. There's Sun (who seem to have a fair chunk taken out by Intel as of late), IBM and so on. So, Intel vs. AMD is not going to be it for servers. It'll be Intel vs AMD vs IBM vs SUN vs etc...

But that's a small thing. I'm not going to discuss at this point as to "Does Hammer stand well as an alternative to Itanium" - I apologize, this at the moment I just don't have the breath for - the post is long enough as it is .

My biggest concern for AMD's Hammer in the server space is VALIDATION. Intel spends HUGE amounts of money, time & engineering resource on validation (as do the other big iron players, but I am merely concentrating on Intel as most will know about them ). That is not only "because they can" (though that helps), but also because the HAVE to do that.

If you're talking serious servers (4-way and up, often 2-way for a fair few people), the "king" is *NOT* performance, but RELIABILITY. And part of this, is validation. I am not saying that I think Hammer will be unreliable. What I am concerned about is that AMD just does not have (again) the financial resources that the other players (primariliy Intel for this article) have. Also, Intel & co. have been in the Server business for years and also make their own boards.

"Just selling the CPU" without having direct involvement with the boards seems a bit risky to me, certainly if you want to get to the higher (and more lucrative) echolons of the server arena. Well - that's my fear at any rate. Also, if you spend 30+ K $'s on a server (can quite easily get that high with 4-way and up systems) do you REALLY want to spend that money on "unproven" stuff. It'd be quite a gamble and I certainly would feel uneasy if I were to spend this much money on someone who's "new" (well - in the "high-end" at any rate) to the market & thus unproven. I think AMD will not have an easy time here.


IN CONCLUSION
*************

Right - so these are some of my fears/concerns & I just wanted to have other people's input on the matter. I think AMD's tactic of "putting all it's eggs into a single basket" is a receipe for disaster. The 64-bitiness slapped on to Hammer is rather futile in most cases IMHO, and their bold (and, I suppose necessary) move into the higher server area will not be without pitfalls. Remember, if margins in something is high, so is the risk. If your 30K+ baby goes down due to some reason (fried CPU's, mobo problem, or whatever), you're not going to be a very happy customer.

Especially because these heavy-duty babies don't tend to be used to have the latest version of Pong or Solitaire played on them, but big business parts hanging off them. And that translates to a lot of man-hours lost for each server downtime. Hence, Server validation etc. is by NO MEANS a "small undertaking" - and I doubt very much so that AMD just has the funds for that. And I've the suspicion that they'll come out of it (if at all - and I hope the don't go down for the count in the process ) they'll have a fair few bruises.

If anyone care to comment on this, please do so freely. However, don't flame please and give reasons to arguments. "Hammer will sell - don't worry." just doesn't cut it with me. .

Thanks for any oppinions.

- Shathal.

 

[mechBgon]

I'm sorry, but the wording is vague enough that your poll needs to add another option: "I cannot understand the other options so I'm choosing this one!"

Let's just put it this way: Intel will be following AMD's scheme and releasing a hybrid 32/64-bit CPU, the Nehalem, in H2 '04. Looks like that's where we're going, necessary or not, and AMD happens to be the one who will do it first.

Not a lot is known about the successor the whole Pentium 4 family code-named Nehalem. It is said that it will greatly expand the idea of the Hyper-Threading (see the news-story about Intel?s Modular processors), will feature the hardware security LaGrande technology and, as stated by the source, the Yamhill 64-bit extensions. It will still remain to be IA32, but with loads of architectural innovations, I believe. In the second half of 2004 the Nehalem will be made using 90nm manufacturing process, but in late 2005 or early 2006, the novelty will be transferred to thinner 65nm technology.

From a source that AnandTech's Brandon Hill calls highly reliable (namely, XBit Labs). XBit's article

 

[Wolfsraider]

my way of thinking is its best not to put all your "eggs" in one basket.
amd isn't so no worries.

people tend to buy into the fear factor too much amd wouldn't be selling processors so cheap unless they made a profit and if they are not making a profit then nothing you or i do will make that much of a difference.

but i highly doubt that amd isn't well aware of their prospects.

times are lean and the industry has to conform / abide by the actions of the consumer to make a profit but amd has quality parts available now and are selling them.

the hammer is the next logical step in the computing world.although you might remember the average consumer won't buy it at the time it becomes available

<-----not saying we are average lol but most folks i know are still buying computers from the likes of rent a center where you can buy a celeron 128mb memory cheap components for a grand payoff of 3000.00 just because they didn't do their homework.
the p-4 and athlon xp have become household words and as such they will be bought on name alone,hammer will get to this status but it took about a year and a half? for thr p-4.so amd will be riding the storm on the products it has available more so than
on what it will have.

my humble opinion
mike

 

[shathal]

Originally posted by: mechBgon
I'm sorry, but the wording is vague enough that your poll needs to add another option: "I cannot understand the other options so I'm choosing this one!"

Let's just put it this way: Intel will be following AMD's scheme and releasing a hybrid 32/64-bit CPU, the Nehalem, in H2 '04. Looks like that's where we're going, necessary or not, and AMD happens to be the one who will do it first.

Well - actually, it won't. Otellini said at times now, that Intel will not be doing a 32-bit/64-bit hybrid, as far as I know. Please update me, if I've missed some confirmation by Intel. I've not been informed one way or another that Intel *DO* want to do it, and I'd be quite concerned if such a thing were to happen & I wouldn't get notified .

Here's some of my references that I searched for on the quick... :

http://www.theinquirer.net/?article=5627

http://www.theinquirer.net/?article=4016

- Shathal .

 

[SiGm0]

Don't most of the hype of the Hammer come from AMD Fanboys

From what I've read the Hammer sounds very nice but also very risky at AMD's current losing state

However if they do manage to implement some of the Hammer's unique features (64 bit course, S-O-I, Memory Controller and so on) they are looking at a very excellent CPU, if they do manage to pull it off then kudos to them

But if they fail as you said with M$ and all other software companies they could be a lot of financial problems, I firmly beleive they should have a backup plan but as we all know the back up plan is never as good as the first intended plan so it could possibly not even help.

Going into 64bit computing was inevitably going to hit Desktop computers, why not later when you can do it now? Of course if it's costing them too much maybe they should do it later but 64bit will be very interesting to see in action. Anyway the Hammer's rivalry from Intel (the Prescott core) is having 64bit too and is coming out shortly after the Hammer.

Although theses delays of the Hammer and Barton are not helping AMD's financial status as each Quater they are losing more and more money they should really release the Hammer as soon as possible never too good holding it too long.

I think AMD knows what they are doing and putting all there eggs in the basket seems like what they are doing but ultimately they should know the risks involved.

We'll just have to see how the Hammer's perform when they hit pc shops

Just my Opinion

 

[mechBgon]

Not a lot is known about the successor the whole Pentium 4 family code-named Nehalem. It is said that it will greatly expand the idea of the Hyper-Threading (see the news-story about Intel?s Modular processors), will feature the hardware security LaGrande technology and, as stated by the source, the Yamhill 64-bit extensions. It will still remain to be IA32, but with loads of architectural innovations, I believe. In the second half of 2004 the Nehalem will be made using 90nm manufacturing process, but in late 2005 or early 2006, the novelty will be transferred to thinner 65nm technology.

From a source that AnandTech's Brandon Hill calls highly reliable (namely, XBit Labs).

I really don't see what you're worrying about in the first place. You're worried that the Hammer includes 64-bit support which might not really be needed? Is there something wrong with including extra capabilities which may not get used much...? And Microsoft has certainly committed to 64-bit Windows, there is evidently already 64-bit Linux, and Sun Microsystems themselves is considering putting "AMD Inside" of some Sun-branded servers. I don't think your anxieties are necessary. Anyway, carry on now...

 

[shathal]

Let's deal with Nehalem, once that's a bit closer at the door. At the moment, Prescott is doing quite nicely - and that has not got 64-bit extensions of any kind that I am aware of (Where'd you heard one, SiGm0 - any sites per chance? ).

Nehalem is a different tin of fish in various ways.

1: It's from Intel, rather than AMD & Intel don't have the "all eggs in one basket" state at the moment, nor would I assume that they get into that stage (it'd be quite unlike them).

2: Intel DOES have the necessary cashflow & industry muscle to bring 64-bit things out.

3: My concern is essentially AMD disappointing people. 64-bit at the moment is just pointless for most, as well as unusable (unless some IHV's and MS are actually writing a 64-bit OS for AMD).

REMEMBER: AMD's 64-bitness stuff is NOT compatible with Intel's 64-bit stuff. So they WILL be going seperate ways, and that's a split where I would feel a little uncomfortable in AMD's shoes (when it comes to wooing Microsoft). Other OS's - yes they may play a role, but by simple numbers (and End-user usability), it's Microsoft that counts the most.

My suspicion for future events is not only that the 64-bit that AMD's marketing dept. is building will deflate, but rather blow up into their faces, and thus cost them even more. And the problem with THAT could well mean that AMD loose out even more. I've not heard of a backup strategy they have *IF* Hammer fails (I suppose someone the size of Intel doesn't worry about this sort of thing, they can just push things into the market). That's what I don't like. The prospect of (potentially) seeing AMD going down the sewers.

I don't personally care much for the processors (just personal preference), I am merely interested in having them around as a competitor to Intel, an alternative that people want and something that keeps Intel (and vice versa) on its toes with product quality, features and pricing, in the end. (See - all selfish reasons ).

I suppose the paranoia scenario I am brooding on, is in effect - will Hammer be the axe that will cost AMD it's head? The concern is not for 64-bitness (IMHO, it's just too early to make any sense) - we will be going there sometime EVENTUALLY .

Hope that makes it clearer.

- Shathal.

 

[SiGm0]

I stand corrected

 

[AndyHui]

You can quite happily go up to 64 GB of memory based on 32-bit stuff (with a certain bit whose name I forgot just now)

PSE36 - Page Size Extension 36

 

[mechBgon]

If you rummage around the Inquirer some more, you will find that Microsoft has gone on record with official support for AMD's 64-bit processors. With a company like Sun considering adopting their own competitor's CPU, I'd have to guess it's got potential. If Intel's next all-new CPU design will feature AMD-compatible "Yamhill" x86-64 instructions, then that pretty much pulls the rug out from under anyone thinking that AMD is going down a dead end, doesn't it?

Frankly, I'm expecting a very concrete development in 64-bit apps for myself: Caligari trueSpace 7, or whatever they decide to call their next-gen trueSpace product. Couple of Clawpterons, a dual board, a stack of DDR, and I'll be a mad l33t rendering fool, y0

 

[Duvie]

MECHBGON IS RIGHT...I also heard of the ms confirmation....


My concern....


Since they can't seem to get chips out now....either it be rumored 50% yields even on their new tbred rev b's which are no near as complexed as the hammer, or that the lack of manufacturing ability...The fact is they have four outstanding chips reviewed out now that ARE NOT available in masses to the ppl....


I think 4-6 months is a dream for the hammer debut...It wont be out in my opinion until July-Aug of 2003....and that is if they don't have more manfacturing problems. I know they haven't said it is having manufacturing problems with the hammer, but I have heard of low yields in trying to implement SOI technology.....


I hope they can just get the barton ready to go which is a minor revision of current chips and solve their manufaturing issues with them (tbreds)...The hammer and 64bit technology doesn't need to come out yet. They are staying competitive "ON PAPER" so far so no need to rush something out that may have problems or before the process has been truly perfected...

 

[imgod2u]

Hammer is a good temporary solution for maybe 1-2 years. But they're gonna have to come up with something better than yet another x86 processor in the long run.

 

[shathal]

One thing I actually remembered that will be also interesting, in some ways, is how AMD will "live" (I'm sure they'll cope) without Intel's compilers.

At the moment, even AMD's benchmarks use Intel's optimized compilers (methinks Intel isn't allowed to put "anti"-AMD stuff into their compilers, or they'd get grilled REALLY badly). I just remembered, 'cos I've seen the sort of difference that using them can make on certain systems / applications.

The funny thing is that they actually help AMD a FAIR chunk as well.

The "problem" is (quite ironical), that now that AMD is doing it's own 64-bit thing, it can't rely on Intel's compilers now, as Intel's 64-bit architechture is radically different and Intel's 32-bit architechture is nothing like Hammer.

It's an interesting side-observation and I am curious to see AMD's next set of benchmarks for their Hammers. Who knows - maybe they finally go about writing their own compilers?

- Shathal.

 

[Cerb]

The fact that Hammer is a 64-bit CPU is marketting. Yes, it is 64-bit, but I'd be quite suprised if it isn't dead by the time too many products supports it. Kinda like the K6-2s w/ 3D-Now!...now all kinds of things use it, but not until the Athlon did more than a small amount of apps. However, the question shouldn't be about the 64-bit abilities as much as how its design, possibly the 64-bit parts even, will help it perform up to snuff with the P4s at 3+ GHz. If they can manage that with other current software and hardware, I won't worry myself about not using the neato 64-bit goodies in it.

 

[zephyrprime]

My primary concern with hammer has always been that it only extended the pipeline by 20% over the existing 10 stage pipeline in the athlon. The Hammer just won't be able to compete on MHz terms with the P4 with only 12 stages.

 

[Pink0]

My primary concern with hammer has always been that it only extended the pipeline by 20% over the existing 10 stage pipeline in the athlon. The Hammer just won't be able to compete on MHz terms with the P4 with only 12 stages.

I'm sorry but I'm having trouble understanding what you're trying to say here. The P4 doesn't have 12 stages. It has 20. The hammer will still be a much more efficient CPU than the P4. In fact, early estimates point to it being 20% more efficient than the current athlons clock for clock due to architectural changes, new cache system, on die memory controller etc. Don't forget that most of the P4's speed comes from SSE2 optomizations and this is how it's able to beat the current athlon xp in most benchmarks. The hammer will have SSE2 also and will level to playing field.

 

[shathal]

My 2p on the latter, if I may...

What zephyrprime was referring to was not that P4 has a 12-stage pipe (it's got a 20-stage pipe, that's true) - it's hammer that has a 12-stage pipe.

Now, while a shorter pipe does allow for technically more IPC (hence, MHz for MHz Athlon is usually better off than P4P. If you don't use PR numbers, a 1600 MHz Athlon will beat a 1600 MHz P4P mostly, this is not the point however).

What a high-count-stage pipe DOES do for you, is that it allows you a really high clockspeed. The longer the pipe, the higher you can ramp your clockspeed, in essence.

True, you have to be careful that your branchprediction algorithms are up to par, as otherwise you have to flush the pipe & begin anew, but that's always the case. It just affects a deeper pipe more than it affects a shallower one.

So, Hammer (again) won't be able to compete with P4P MHz for MHz. IPC can't be used as "the ultimate" performance measurement tool (purists may likely disagree). No matter how you spin it, a 2.8 GHz P4 (and soon - 3.06 with HT - slurp) does beat pretty much every Athlon out there in pretty much all benches. Admittedly, the Athlon (due to a shorter pipe) has got a better IPC performance, but with "so many more cycles" the P4P still beats it. That's one of the reasons I don't see P7 (or as Intel Mktg call it - "Netburst") as a bad architechture.

True, a purist in me would also prefer to see "lower MHz, More IPC", but as long as the performance is right in the end, who cares?

With just a 12-stage pipe, I'm curious to see how well AMD can ramp hammer. In theory, it being a new architechture, Hammer should ramp reasonably well - as most new architechtures do. The problem I see is that with a 12-stage pipe, they may not be able to ramp fast enough ...

- Shathal

 

[shathal]

Urm - minor correction to my post above. What I meant to say is that my CONCERN is that Hammer (again) won't be able to compete with P4 in the "long" run, as P4's clockspeed will just go up & up (not seen signs yet of it stopping). My suspicion is merely that Hammer's performance won't be able to scale as quickly as P4's that's all in this case.

Just paranoia .

- Shathal.

 

[Degenerate]

The delay is certainly not helping.
was due last quater about?, then moved to this quater, then moved to the 1Q of 03. Then, just read at the inquirer, expected at 1h of 03. the way this is going is not exactly what AMD is hoping for, regardelss of the product.

anyway, i think if they can make it a good CPU, then they still can be competative, trouble is, if they can't make it good (performance, and now the delay is coming in effect), then It's hard to imagine what AMD can pull next.

 

[shathal]

Yeah ... the way things are going at the moment, Hammer will be competetin with Prestonia (which will have More L2 cache, faster FSB, code optimisations and new u-code instructions AFAIK), rather than "just" Northwood with HT in it.

It's a hard call to make. If they DO have product issues (which I assume is the reason for the pushbacks?), then obviously need to sort them out. The longer they wait though, the "more" lead to break even or even gallop ahead they give to Intel. Not an nice spot either way really.

Ah well. We'll see "soon enough" (well - 6+ months) if AMD can deliver on their promises (to deliver product) and performance-related promises as well.

Different question entirely is how well the CPU will be taken up in the server (part. the "real, serious" server) market. Not really something where you can feel your way in, rather have to hop into the deep end right away. And THAT (though necessary, I think, not really a gentler way I can think of) is bound to cause some nasty "eggs in the face".

- Shathal .

 

[Sheddd]

PSE36 - Page Size Extension 36

Are slow and inefficient; from what I've read it beats a raid-array's speed by ~10%. Ouch. Emulating 64-bit addressing just doesn't work very well (at least not the way Intel's done it).

Urm - minor correction to my post above. What I meant to say is that my CONCERN is that Hammer (again) won't be able to compete with P4 in the "long" run, as P4's clockspeed will just go up & up (not seen signs yet of it stopping). My suspicion is merely that Hammer's performance won't be able to scale as quickly as P4's that's all in this case.

The P4 is Intel's shittiest design ever; If we had PIII's at .13 micron it'd be whipping the P4 on benchmarks like a red headed stepchild (and you'd be getting lots more per wafer).

I'm hoping Hammer will be a big win for AMD in the server market... otherwise I think AMD's doomed and we don't want that. Remember what processors used to cost before AMD was compeditive?

 

[naukkis]

The P4 is Intel's shittiest design ever; If we had PIII's at .13 micron it'd be whipping the P4 on benchmarks like a red headed stepchild (and you'd be getting lots more per wafer).

We sure have 0.13 P3's and they aren't whipping P4's in benchmarks, actually they are so low-performing chips that nobady don't even care about them.

 

[Wingznut]

Originally posted by: Sheddd
Remember what processors used to cost before AMD was compeditive?

I'm not saying that competition doesn't help lower prices... As it certainly does. But market conditions have an even bigger impact.

For instance, things were never more competitive than the race to 1ghz. Any guesses as to how much an AMD 1ghz Athlon cost?

"AMD is currently shipping its 1GHz AMD Athlon processors priced at $1,299 in 1,000 unit quantities." --AMD

If processor sales were as high as they were two years ago, both AMD and Intel would be selling their products at a higher price.

 

[bfonnes]

If we had PIII's at .13 micron it'd be whipping the P4 on benchmarks like a red headed stepchild (and you'd be getting lots more per wafer).

.13 micron PIII = Tualatin

Bfonnes