Why not AMD? (aka, Apple and x86 Questions, Part 4)

Do you think Apple will eventually move to AMD? They are by far more innovative than Intel, who is reacting to AMD and not leading the market. AMD is eating Intel's lunch! I could deal with moving from the PowerPC if they were using the best available, but Intel is the lowest denominator.

A lot of Mac users are really hung up on the Apple choosing Intel over AMD thing, but to really go into this we need a clearer picture of AMD's history in the microprocessor arena and their current role in the marketplace.

Right now, some of the antics you see going can and have really muddied the situation in enough people's heads that the way they view AMD is vastly different from the reality of their situation. Note I said clearer, as while AMD does have an interesting story, and we'll be going into some of the differences in their roadmaps and differences between their products and such, this isn't magnifying glass as this isn't titled 'Deconstructing AMD'.

I mentioned the reality of the situation, and this can get tricky, but think of it in terms of something's brandshare and mindshare in disparate groups versus what's really going on. Apple is actually a really great example of this, as several years ago I was having a conversation with someone very PC centric about some reseller stuff, and he said, "You know, for such a small company, they sure do make a lot of noise."

Now, if you're an Apple-head, that can throw you, as in your head the market is Apple and Microsoft, but I hate to break it to you: Apple is barely on Microsoft's radar anymore in a real way. They aren't worried about them in the same way they worry about say, Linux. Apple has been reasonably lucky here, because Microsoft mostly just ignores them instead of firing back (They do have some ammo) because there isn't a whole in it for them.

Of course perception is reality, which is one of the reasons why you see Apple trying so hard to sway perception when it comes to their standing in the marketplace, and one of the reasons why Apple seems to pretend Linux doesn't exist. 'Unix' exists when it comes to bringing software over, but not 'Linux', because Apple can't afford for the story to become all about Microsoft and Linux as third wheels have a hard time keeping mindshare.

Anywho, the reality is that Apple's world-wide marketshare (I.E., their percentage of new machines sold) is under 3% in a big way, and depending on who you ask, is actually well under 2%. These don't take into account the Mac Mini which was introduced, but even when those numbers are out I doubt Apple will be much above 4%, if they even get that. There are people predicting things like 6% by the end of 2005, and let's just say I'd be astonished.

As a contrast, the rest looks something like this:

  • Dell: ~34%
  • HP: ~20%
  • Gateway: ~8%

Of course there are people like Sony and all who have their shares, but when you add them up you're left with a weird sum: almost 50% of the market is missing. These are the white-box computer makers, and run the gamut from those guys who have a shop on the corner and build your PC for you on the low-end, and the people putting together custom computers like Alienware.

Stop and think about that for a minute, and you can start to get an idea of the economies of scale the x86 processors have over something like the PowerPC when it comes to personal computer, and just how big of a sea Apple's boat is sailing in. At this point for Apple, it's about carving out a self-sustaining island while it tries to branch out and grow, mostly in niches. The desktop wars ended long ago, and everyone has moved on to new battlefields.

Dare to dream and all, but those are the numbers, and most people know it in their gut anyways, which is why so many are really freaked out about Apple and the Mac right now since the iPod seems to have their majority focus, and why there has been this impending sense of waiting for the other shoe to drop' in the community.

Still, while the desktop wars ended long ago, you often wouldn't know it by how Apple is talked about in the press, or how a lot of fans talk about new developments. This is generally a conglomeration of:

  1. The media loves a good horse race. They love tearing things down and building them up, and while there is often a glimmer of truth in the process of both, you can't really fit the context of a situation in a sound byte. Yeah, I know, this is the domain of politics and celebrity, but it takes place in technology circles too, and something like FireFox might be a great example also.

    I love Firefox, and they've made strides, but it isn't as though Microsoft has them breathing down their neck to the point of game over... but you wouldn't know it from some of the stories out there.

  2. Opera's share of the browser market isn't very far behind Firefox's, but if you've noticed, if Opera lit itself on fire, the story would be how Firefox heard Opera lit itself on fire. [note: A reader passed on an article in the comments that shows Firefox may have started to pull away from the pack in a big way in the last 6 months ]

    Now, Firefox getting publicity is good for them in a way, because it at least puts the idea of an alternative as an option in people's minds, but the point is that stories about Apple just sell, because people are passionate about the brand and curious. It's the same thing that gets Apple on the covers of magazines.

  3. Hardcore Apple fans, like hardcore fans everywhere, are functionally retarded when it comes to the object of their obsession. They can't help it, the object of their obsession has such a large place in their head that they become just a little... myopic.

    I don't really fault them for this, because I'm the same way when someone says they love the Pixies, but only really know "that one song from Fight Club". I lock the door and start breaking out the bootlegs while going on about being in High School and hearing Doolittle for the first time while Jane's Addiction was playing in the backgraound and hearing them slowly meld into a message from God that this music was for me.

    The point is, the Pixies take up a ton of headspace for me, but only a little bit for most of the general public. I don't think there is anything wrong with that, but it needs to be accounted for, because I'm hardcore about them, just as it needs to be accounted for here.

    Where it can get a little sad is that obsession and passion can so easily be manipulated and/or abused, as it's not as though all of those "Anti-Mac" articles are really written because the author thinks it is the truth, and while one side may be 'faithful' that doesn't account for what the other side is up to.

We're going to break out of that list and jump to ignorance, because, well, it's the natural evolution. There's nothing really wrong with ignorance per se, it's generally just a state of unknowing, and everyone has ignorance patches... usually around things they don't have to deal with on a daily basis. I don't expect Joe Public to understand the finer points of obsession, just as Joe Public knows it will take me two hours to change the tire on a car.

Now, let's flip it and think about how most people think about Macs. For the most part, they're entirely ignorant, and if asked about them might say:

  • Apple makes computers?

  • Yeah, I've used Macs. My school had them... those black and white things?

  • Yeah, I've used Macs, my school has some old iMacs down in the lounge that no one uses.

  • Aren't they supposed to be really good at graphics? And music?

  • I've never seen one, but I've heard they're really nice.

  • Macs don't have games, do they?

  • Oh yeah, those. Expensive, and slow. Kind of toys, really.

  • etc., etc., etc...

A lot of Mac users will then either:

  • Take this really personally, as obviously the person they're talking to doesn't know technology at all if they don't have a flying clue about Macs.

  • Use it as an opportunity to be incredibly annoying and deign to make the person sit and listen about just how great Macs are, and how stupid the other person was for buying a PC. I know, because I've been there with the Pixies, and back in the day, may have done this regarding something called Rhapsody that never really shipped. (I was never really sold on the Mac to the point of zealotry, but I'll admit to buying into the Rhapsody promises in a huge way)

Most people generally just don't deal with Macs on a daily basis, or even encounter them, so they have little snapshot opinions. Impressions they've picked up along the way, but haven't really had changed. Macs often are slow for long periods of time, and when they become faster than what's out there, they need to be faster for a long time to really counteract the mindshare that's out there.

Anywho, this brings us back to AMD, and the impressions a lot of Mac users have of them, and the impression a lot of geeks have of them too. It's not necessarily a wrong impression, it's just incomplete, which can end up being misleading.

Historically, AMD has been a bottom feeder. They really were a follower, nipping at Intel's heels with lame-but-very-cheap-cores and horribly cheap and buggy supporting chipsets. Basically, for ages and ages, the joke would be you bought a machine with Intel inside if you couldn't afford a Mac, and you bought a machine with AMD inside if you couldn't afford Intel.

Chipsets are kinda what props the CPU up. A CPU is pretty boring on its own, and you need things like a logic board to make it interesting, which means you need things to allow it to communicate with RAM, ports, etc. A major problem with AMD was that they outsourced the making of their chipsets to others, generally overseas, and if you were putting together a cheap system in a market where margins are already razor-thin, if you can cut a corner somewhere and stuff should still work, that corner is going to get cut.

Of course most of those things are expected to be there, and if you were using an AMD chipset, things had a tendancy to go wonky. It kinda became a catch-all blame-funnel if you asked someone about problems you were having, similar to "repair permissions" in Mac OS X.

To hit this home, historically AMD has not been associated with:

  • Innovation

  • Quality

  • Performance

Now, through a variety of means (much of which amounted to them taking things into their own hands) the quality of the supporting chipsets has improved, as has the performance of their chips, although the price has gone up. However, before the race to 1 GHz, most Mac users had little idea of who or what AMD was, and that probably started to color their perceptions a bit, as did Apple's bashing of the Celeron for quite a long time.

If you'll recall, at the time the MHz on the Mac was going nowhere fast -- or slow -- and just about everything tech revolved around this AMD and their Athlon in a good old fashioned horse race with Intel for the race to 1GHz... and that horse race cost AMD gobs of cash but got them gobs of mindshare. Perception and PR really do matter, and while Intel's business didn't suddenly go into the toilet, they of course responded (and perhaps overreacted) and made sure they'd be king of the GHz crown for a long time to come with their future offerings.

All of a sudden, AMD chips were fast, and for what you got in performance, cheaper than hell compared to Intel's chips. It wasn't as though AMD had some magical process for making them super cheap though, much of it just came from the fact that AMD only rarely actually turns a profit.

This is actually something worth bringing up, as it's not something a lot of Mac people are aware of... but AMD is not the healthiest company around, and has gone through periods, even recent periods, where its operating cashflow was very shallow. It's got debt, lots of debt. It isn't very diversified, focusing primarily on its core microprocessor business and flash memory. [note: A reader pointed me to this article, which I'd missed, saying AMD may well be dumping its flash memory business altogether]

You know, the stuff used in cell phones and cameras and such, which is just another of the areas Intel has a major stake in. Another area might be ethernet controller chips, which, you know, Apple has even bought and included in its products.

A good quarter for AMD is often one where it beats expectations yet still takes a loss, and to a certain extent, it just has a problem winning as when one of the areas it is focusing in is doing well, the other is just getting hammered.

An example might be when it was getting hammered in its CPU division, yet flash memory at the time was hot hot hot and AMD was turning a profit because of it, yet in the last quarter of 2004 (I haven't looked at Q1 of 2005 beyond marketshare stats) their processor division was turning a profit (Yay Opteron!) yet its flash division was getting positively hammered due to Intel deciding it, you know, wanted to just hammer them there.

Now, when AMD took an operating profit in its CPU division for Q4 of 2004 -- a huge quarter for it -- the division took in around 80 million in profit. In the same quarter, Intel was bringing in profits of over two billion. The specifics of this are fuzzy in my mind, but I do believe Intel was part of AMDs problem in the flash market, as they decided to compete very rigorously, and there are always rumors floating about that Intel plays cat-and-mouse a bit with AMD just to keep the anti-trust people off of its back.

At the end of the day though, all you need to know is Intel grew its business in NOR flash memory by 27% and AMD's memory group lost 20% of its marketshare.

This is an aspect of the companies that really can't be underestimated too much, considering the areas they are playing in, which are very, very expensive games:

  • Intel has a habit of just raking in the cash, year over year, and when you are pulling in ~$7+ billion in profit and your competitor isn't pulling in profit year over year, almost ever, it's going to take a toll. That cash allows you to do things your competitor just can't, which means you have options. You can correct course without going out of business. You can funnel extra cash into R&D problems that would cause your competitor to run in the red. You can give more resources to multiple battlefronts.

  • At this point in the game, one of the scariest aspects of microprocessor technology (aside from its environmental impact) is its exhorbinant costs. There are neat areas of the market where you can fire off a company your design and have it pumped out, but the smaller you go the more exponentially expensive the process technology (and research) becomes.

  • Any fabrication plant in general has astronomical upfront costs, even if its just pumping out older 130nm+ tech. We're talking billions. These things don't get built every day, but you really need them if you are going to be able to provide the volume necessary to play.

What you should take away from some of what I'm hitting you isn't that AMD is about to die, but rather that AMD is not in the healthiest position in the world and sometimes what looks like a lead or a breather is really the time between punches.

They do have a new fabrication plant close to coming online in Germany, which should be interesting, and they've had a technology-sharing agreement going on with IBM for quite awhile (and which has been extended for several more years into the future) which has helped them big time when it comes to research and development.

And of course, you have to give AMD some credit here too. One could make the argument they've steered the direction of the entire market twice in recent history, while having well less than 20% of the market. (Just to note, don't think of AMD as being 15-20% of the size of Intel, as you have to remember just how many pies Intel has its fingers in; everything from graphics cards to controller chips to PDA CPUs to flash memory.)

AMD is also the underdog, and Mac users, and some would say the entire American culture, seems to be able to identify with underdogs and loves to root for them.

Mac users have heard the name HyperTransport, because it's in G5-based Macs, and AMD was really the key player in the technology and has pushed it in their 64-bit lines. They doubled-down with a huge bet on extending x86 rather than reinventing it, which has paid off in spades. IBM and others have multi-core chips in their very high-end offerings for awhile, but here was AMD raising the pirate flag and trying to bring it to the masses while carefully derailing Intels nicely-engraved roadmap.

It's really hard not to be charmed by AMD, especially when you see the performance of a decked out Opteron-based server compared to anything else on the market, including an XServe.

The Opteron in particular shipped (awhile ago) with an integrated memory controller, which puts the memory controller on the same die as the CPU, meaning you have them as one package. This is so kick-ass, as every time the CPU needs to communicate with RAM it has to go out over the system bus to communicate with the memory controller, which is sucking up bandwidth on the bus that can't be used for other things.

With an integrated controller, it's right there on the chip, so it can do more with the bandwidth available to it on the bus and access memory that much faster. It really has a dramatic, and will continue to have more and more as more CPUs are accessing memory, more memory is being put into machines, and CPUs are needing to be fed faster and faster. It's a full generational leap beyond what the PowerPC 970 or Intel's offerings have, and AMD has been pumping it out for years.

One really has to give them their due here.

Back in the K6 days the idea of Sun, Hewlett Packard, and even IBM building much of their kit around their offerings would have been laughable. They've come a long way from their bottom-feeding days, even as their tech has become more expensive.

And yes, an Opteron-based XServe running Linux and even Mac OS X would certainly rock. There's a reason why Apple is hot to show benchmarks with Photoshop/etc. on the PowerMac G5 page, yet there's nothing showing how the XServe stacks up for server-tasks.

Trust me, you don't want to see it. They're brutal, and it doesn't just have to do with the fact that there aren't any quad or 8-way XServes, let alone blade systems. While some of it has to do with OS X's architecture, a lot of it is just that the Opteron kicks ass at these tasks, and it's not just Apple feeling the heat. The server space is really the only area AMD has shown some real gains in since the Opteron shipped.

At the same time, you can't forget about where they really are in the marketplace and the heat they are under.

AMD recently filed an anti-trust lawsuit against Intel, and in their CEO's usual almost-bizarre way started posting open letters on their website (nothing against almost-bizarre, but it's amusing to hear a CEO say something like "it was a freakin' dismal quarter in flash [memory]", and in one of the PR statements surrounding it said:

Intel’s share of this critical market currently counts for about 80 percent of worldwide sales by unit volume and 90 percent by revenue, giving it entrenched monopoly ownership and super-dominant market power.

One of the reasons the above is great marketing is that it encourages you to adopt the soundite and think of the market as being Intel and AMD, with AMD having 20% of the market, whereas it is more like 16%, but bygones, the revenue and marketshare mismatch should be enough to raise your eyebrow.

AMD certainly has some valid gripes against Intel, and would love to see them thrown under some serious anti-trust regulation, but a lot of it is just a PR stunt as AMD knows the deal. Even though many people think of them as "eating Intel's lunch", they have had very little marketshare movement. In fact, they're about exactly where they were back in 2000:

  • 2000: 16.7%
  • 2001: 20.2%
  • Q1 2002: 18.2%
  • Q4 2002: 13.8%
  • Q1 2003: 16.6%
  • Q4 2003: 15.5%
  • Q1 2004: 15.1%
  • Q4 2004: 16.6%
  • 1Q 2005: 16.9%

The one area where AMD has seen some real movement, as I mentioned, has been in servers, but even there it hasn't been storming the beach. Back in 2004, AMD was saying they expected to have a marketshare in the server space of 10-12%, and by mid-way through 2005 they still aren't close, knocking in at ~5.7%, which is up from the previous 2004 quarter (5.4%), but...

In the same quarter that AMD sold ~92k Opteron systems (Q1 2005), there were ~800k Xeon systems sold, and the majority of Opteron systems are still sold via little boutique and white box shops. When you look at a lot of the numbers, the clearest thing that jumps out at you is that the server market is the only place they're seeing real movement, and even there it is barely a threat to Intel right now in the hard numbers, it is all a perception thing. While AMD's server market share before the Opteron was basically a rounding error, Intel is still raking it in.

AMD is seeing some boosted sales in the general consumer market, but they are primarily coming from existing AMD owners, and we're brought around to another point: When something has become the defacto standard, as Intel has, it's not enough to be better to get people to switch, you have to orders of magnitude better or it's just not worth the hassle. Or, you can be so much cheaper that it is worth someone's while to give you a shot.

Basically, AMD has to be very cut-throat when it comes to pricing (as they have historically) in most of their markets, and the margins in the consumer space, with chips like the Athlon XP or the Sempron are brutally thin even though they've gone up a bit in price as AMD's tech has improved. That very slight server uptick is actually a godsend for AMD, as while their numbers haven't improved drastically, an Opteron is a (by AMD standards) very expensive solution and brings in very juicy margins, as AMD's balance sheet has reflected.

And of course, the Opteron is just one aspect of the market. Let's take a quick look at AMD's lineup going forward:

  • Servers and workstations

    The Opteron going to get the most attention in the second half of 2005, with three dual-core versions sliding into the market and starting to push away from the current socket they're using. AMD's initial dual-core solutions used the same package and pins as prior Opterons, so you could basically buy a dual-core chip, flash your BIOS, and be good to go. Great for AMD, although less-great for those selling AMD solutions, and they're moving towards a different pin layout now. I haven't heard what is going to be going on with it into 2006.

    There is still the Athlon MP, which is basically a version of the single-cored Athlon setup for having multiple processors in a system, but it isn't a major focus and is just sliding along the road map.

  • Desktops and desktop replacements

    This is primarily the domain of the Athlon 64, in its vareties, but also includes the Sempron which is its very pared-down budget chip. AMD is basically focusing on two branches of Athlon 64, the Athlon 64 FX and the Athlon 64 X2, with the FX being higher-clocked but single-core, and the X2 being dual-core but with a lower-clock per core.

    The word is that AMD will look at moving the FX line into dual-core sometime well into 2006, but it has some major problems here and single-cores are going to be around for quite awhile. There is the problem that most games are really just starting to look heavily into threading (due to Intel's hyperthreading becoming so prevalent), with most of them not taking great advantage of dual-cores, but AMD has a much larger problem here which we'll get to shortly.

  • Mobile and notebooks

    AMD's big push here is with the Turion 64, which is a follow up to their Mobile Athlon64 series, and is going to be their major push in this area through 2005 and well into 2006. When compared to say, a Pentium 4, it compares well, but it is not a truly mobile CPU in the vein of a Pentium M which makes up the Centrino line.

    It is a great choice for say, a desktop replacement portable, but you won't have this in anything but a behemoth. To give you an idea, the Mobile Athlon64 generally runs more than three times the wattage of say, the Pentium M. In fact, the Mobile Athlon is basically just a regular Athlon that has been better tested so that they know they can run it at a lower wattage (and performance) than the regular.

    The Turion64 is actually supposed to be a real mobile CPU,, and brings SSE-3 support (Which the Pentium-M doesn't currently have), AMD's 64-bit tech, and an integrated memory controller, which as you've seen me gush about, is a very cool thing. In addition to that integrated controller, the front-side-bus is being speeded up to 1GHz (from 800MHz, and yes, the current G4 doesn't have an integrated memory controller, but has a bus of 167MHz...) while the Pentium-M is hanging out at 400-533Mhz.

There are two real problems here, both with dual-cores and the mobile arena:

  1. AMD's dual core conundrum

    Without a doubt, the high-end is AMD's shining area right now. When it comes to performance in the desktop or server arena, AMD is really garnering the enthusiast eyes because their chips generally just perform better per wattage, per clock, and overrall. Comparing a dual-Opteron system to a dual-Xeon system almost makes one feel a little sick, in the same way as seeing a 90lbs guy getting into the ring with a 200lbs guy when you know the little guy ain't Bruce Lee.

    The Xeon isn't horrifying, but compared to its competition, it is lackluster and behind the Opteron. We'll just leave it at that, and look at where they Intel and AMD both are when it comes to dual-cores, and where they are heading...

    Intel's dual-core offering is the Pentium Extreme Edition 840, or the Pentium D, which tops at at two cores running at 3.2 GHz, 1 MB of L2 cache for each core, with both cores sharing an 800MHz FSB to the system but which will probably soon be ramped up to the 1066 MHz bus the current Pentium4's use. It also includes HyperThreading, which on a P4 can show some dramatic gains, but on a dual-core system can be problematic.

    HyperThreading is cool, and on a single-CPU can show a big boost for a lot of apps, but if you are just running one or two apps which aren't well-multithreaded on a dual-core system the overhead can harm performance a bit. However, if you are using a dual-core chip running a bunch of apps at once, or a bunch of apps that are well threaded, it can really help. Right now, if you were a gamer, you'd probably turn it off via the BIOS, but in a year or so you probably wouldn't.

    Intel's dual-core chips run hot as hell right now compared to AMD's, as, well, their Pentium4 chips were already getting way the hell up there in terms of heat, especially at 3.8 GHz. The good news is that at idle, with two cores the Pentium-D at 3.2 GHz is about as hot as a single hot Pentium4, but it can eek even a little higher when you are going full bore.

    In contrast, AMD's dual-core offerings have fantastic thermals. Stellar, really, as even though each core tops out at 2.4 GHz right now we're looking at a top-end AMD Athlon64 X2 using 60% of the power of a top-end Pentium-D. This is a big deal, and is the type of thing that can let you slice 100W off the power supply making the whole system cheaper and cooler.

    AMD also has many of its technology leads in its dual-core chips, like HyperTransport and 64-bit processing in its dual-core offerings, and there is still that integrated memory controller.

    Running at a much lower clock speed, AMD's Athlon64 X2, in terms of performance, will generally be on par with the Pentium-D and in some cases will pull a bit ahead. Often not by a huge margin, but clock-for-clock, and in performance-per-watt, AMD is smiling a cheshire grin in terms of technology. If only it could get people to buy...

    On the low end there is still a single HyperTransport channel connecting both the cores to the rest of the system, but on the Opteron... the performance is astounding. A dual-core dual Opteron 875 (4 cores) at 2.2 GHz almost keep up with a Quad 3.3 GHz Xeon with 8 MB of L3 cache. It's really just brutal in the server arena for Intel, and I don't know anyone lusting after a big Xeon solution like they are the Opteron, but it is where Intel and AMD break pace on their dual-core offerings.

    Intel actually 'beat' AMD's dual-core launch, but there is no dual-core Xeon, Intel's Pentium-D is primarily geared towards the enthusiast, workstation and personal computing space. In contrast, AMD is pushing dual-cores at the server and workstation market, with a heavy emphasis on the server. AMD says that servers are where it's needed most right now, while Intel says consumers playing MP3s and DIVX movies while downloading in the background and running virus software will get the most benefit, as it is more cost effective for a consumer to pick up a Pentium-D than a full dual-CPU system.

    Who is right? Well, they both are, and of course most systems will be dual-core eventually.

    This is where this gets really weird, and you'll see why I went into all that stuff about fabrication plants earlier: Intel is just stomping on AMD when it comes to dual-core pricing. Compared to the Athlon64 X2, the Pentium-D can be up to half the cost.

    Yes, half the cost. From Intel.

    Performance is comparable, even though it has a higher clock speed, and the Pentium-D runs hotter, but the high-end Pentium-D (two 3.2 GHz cores) costs the same as the low-end Athlon64 X2, and the high-end X2 is over $1k while the Pentium-D tops out at ~$500. The Pentium-D requires a new motherboard, while you can drop many of these first generation X2 into many existing motherboards for a nice boost, but Intel has a massive price differential to play with here, almost a frightening one.

    AMD answer for the price differential is that it's a boutique chip, so it can charge a big premium for it out of the gate, but the truth is they probably don't have much choice in the matter.

    When it comes to fabrication plants, there are two sizes we'll bring up here: the size of the process and the size of the wafer. Chips are made via a souped-up lithography process, where you basically 'project' the lines of the chip onto a circle-shaped layered wafer which are then etched chemically, and finally the chips are cut into individual units. Once you've built the fab and installed all the tech, you're basically printing cash, with the cost of the wafer being your big cost, and of course that wafer is expensive.

    The smaller the process, the finer the lines of the chip can be, and the more chips you can fit onto a wafer. The bigger the wafer, the more chips you can fit onto it using a given process. Obviously, this ties directly into how expensive each chip will be. Since the wafer is your fixed cost, once you've made back your research and development and equipment costs, the amount of chips you're getting out of each one is directly proportional to how much they will cost and how much you will make.

    All things being equal, the same chip going out on a 130nm process, and a 200mm wafer, will be much more expensive than the same chip using a 90nm process on a 300mm wafer, and this is really becoming AMD's achilles heel. 300mm doesn't sound that much larger than 200mm, but when it comes to surface area for chips, you're talking an increase of 225% over 200mm. That's a huge deal, and years ago Intel decided it didn't want to get short and started investing billions into fabrication plants and the technology needed to improve processes.

    Intel has had 300mm running for several years, while AMD basically had to go around looking for partners to help it keep up in process technology, because when Intel can just decide to drop a few billion and build a plant, and its research and development budget is over 7 times that of AMD's, it knows it has a problem. AMD countered by becoming a partner whore, sharing tech with anyone it can in order to get help and spread the R&D dollars around. Motorola, IBM, and even Apple have been partners.

    UMC was a partner, and in the early part of the decade, theoretically they were going to create a joint-plant in Singapore which would be AMD's way of keeping up with 300mm stuff, even if a bit late. It didn't happen, so AMD sidled up with IBM and such to help with capacity and technology, but that gets very expensive very quickly.

    Right now, AMD basically has its one fabrication plant until its plant in Dresden, Germany comes online (hopefully) later this year, and the Dresden plant seems to at least be focusing on 300mm wafers and the upcoming 65mm process. Of course its costing AMD several billion dollars, and it is just going to get worse: From what I've heard, by 2007, new fabs are going to be approaching $7 billion. It can get help from others, but it's really constrained.

    Meanwhile, as mentioned, Intel has been pushing on 300mm for years, and if I recall, back in 2004 dropped several billion dollars to convert one of its older 200mm plants to 300mm, giving it five 300mm plants total.

    Stop and think about how much of the market Intel already supplies, and you can get an idea of just how much more capacity they already have. When you stop and think about just how expensive this stuff is getting, and that companies can't just go around throwing 70% of their operating costs into new fabs every year, you can get an idea of just how much better positioned Intel is going forward.

    AMD probably just doesn't have much of a choice in what it's charging. The Athlon64s aren't cheap chips to start, and with its smaller wafers and outdated tech it's basically having to charge you for two full CPUs anyways, and if they have to go that route, might as well charge you for three. For Intel, due to their fabrication investments, the situation is very different, and could well be so for quite some time.

  2. AMD's mobile headaches

    The Mobile Athlon was based on the 'Clawhammer' Athlon core, and was very very big both because it's a big design and it was produced on a 130nm process. For the Turion64, AMD moved to 90nm technology and was able to cut the die size way down, which is the same process the Pentium-M is currently using. Eventually the Mobile Athlon64 will move to 90nm also, but right now the process shrink allows the Turion to suck down less power and create less heat.

    (except, because it is smaller, that heat is more concentrated and needs to be watched and accounted for in your design, or you are stuck with liquid cooling solutions when you try to push them, like the current PowerMacs)

    The Turion is supposed to be a direct competitor to the Pentium-M, so the Mobile Athlon64 is still sticking around, which is interesting in the sense that if it works well, AMD can keep their share in the desktop replacement crowd with the Mobile Athlon but pick up some more in the actual notebook space which they've been all but drummed out of.

    There are some problems. The Pentium-M runs between 1.6 GHz and 2.13 GHz with ~21-27 Watts, while the Turion64 runs at 1.6 GHz to 2 GHz (the Mobile Athlon tops out at ~2.2 GHz already) with a thermal signature of 25-35 Watts. Out of the gate, on paper, the Pentium-M is still winning in core wattage, but you have to give AMD some credit when you realize the Mobile Athlon64 was running between 62 and 81 Watts. Huge difference there.

    When it comes to power though, that integrated memory controller rears its lovely head again, as AMD gets a nice boost by having to only connect to one chip on the board, whereas the Pentium-M has to have two chips on the motherboard; so its worth taking into account that the Turion64's wattage numbers also incorporate the integrated memory controller.

    Of course, the Pentium-M has some of its own spicy technology that Intel put a bunch of effort into, and the Pentium-M still wins pretty handily, generally besting the Turion64 in power consumption by well over 20% both when under full load or just sitting idle.

    And at these wattage levels, 20% is a big deal in terms of heat. It's a major step forward for AMD, but Intel is still ahead in terms of thermals.

    When it comes to performance, the Turion64 is a little faster clock-for-clock under synthetic tests, but doesn't scale as high as the Pentium-M, which means you can get more performance out of an Intel solution for general tasks, but the difference isn't very large and they compare well. Under non-synthetic tests, like say, firing up a game and checking its frame rates, the Pentium-M pushes ahead by about 20%, although the top-of-the-line Turion64s generally edge out the the Mobile Athlons. That's a pretty large leap in performance, and much of it is due to Intel giving its CPUs more cache (Pentium-M's are also a bit pricier).

    At the end of the day, AMD now has a mobile chip that gives performance about on-par with the mobile Athlon 64, using drastically less power, that also offers 64-bit processing and SSE-3. Pretty cool, and there is a market for it.

    The problem is, the Pentium-M is generally offered as part of Intel's Centrino chipset, which means its bundled with things like Intel's low-power WiFi chips and such, and as anyone who has watched their battery power die while using WiFi, the CPU is just one part of the equation, and AMD is just dead to this market by arriving so very late.

    Centrino has become synonymous with the best solution for mobile computing, and AMD just doesn't have even a toe in the door here anymore, and they are arriving very late to the party with a product with poorer thermals than the Pentium-M, lesser performance under most usage patterns, and the real kicker is that Intel is gearing up for their next generation of the Pentium-M which will most likely be a broadside with all guns firing. Intel is pumping a ton of cash into R&D, and a big chunk of it is going into mobile technology.

    This is particularly problematic for AMD, because while they have something credible now in what is about the fastest growing segment of personal computing, aside from 64-bit and a cost advantage, their big offering is somewhat lackluster. They've been focusing on the high-end performance, but the big kicker for them is whether they have the resources in the future to pump R&D into the mobile space which is going to become more and more important.

    Remember, lower-wattage CPUs aren't just for mobile markets, as their thermal signatures are what are needed for smaller and cooler PCs in general, and at this level, the research and development costs are getting scary.

Something you also have to keep in mind is that when it comes to the portable market in North America, Apple's market share is substantially larger than its overral share: On the order of 5%. Put two and two together, and it means the portable market is really, really important to Apple even compared to its other product lines. Most people know that, and since Intel is the major leader in that space -- they have the market practically locked up due to AMD's late arrival -- Apple really needs those Pentium-M's.

Ok, so if you've been paying attention, you're wondering why they don't use Intel in the mobile space, but use the Opteron or Athlon64* for its higher-end offerings?

As you may have noticed, AMD has yet to really crack anything but the white box market, and even the majority of those putting out Opteron servers are tiny little shops. One of the big reasons is that it really pays to be an all-Intel shop in volume. AMD has some anti-trust rumblings going here, under the idea that Intel is using its 'proposed' monopoly to force manufacturers not to touch AMD, but it's often not that cut and dry.

Intel will spend millions upon millions marketing the Centrino name, but even if your computer uses a Pentium-M, you can't call it Centrino unless it includes the rest of the package, or Intel's supporting chipsets. Intel can just set price breaks at volumes you'll only get if you are only an Intel-only shop, and if you're really pushing Intel's solutions Intel, with its very lined pockets, will chip in heavily towards your marketing costs since they're getting something out of it too.

Some years, Apple has spent over $100 million on marketing alone, and when you consider that many quarters that's more than they bring in in profit, having that offset by say, 50% gives Apple options. They can get more for the same amount, or they can get the same for less, or they can help use it as a buffer while they're transitioning over.

Intel also offers more complete solutions than AMD, which can't really be overlooked. You can get a... complete... solution via AMD, a very good one manufactured by NVIDIA, aka the nForce, but that's about it. For the most part they just throw out reference designs and people around the world cut whatever corners they can to have a cheaper system. You can get just the chips from Intel, but it starts to get a silly, as unless you need something really esoteric, Intel has taken their chipsets and outsourced the making of the boards to others, which means you have their economies of scale and testing behind you.

With Apple making this move, it would be almost insane for them to still be making their own chipsets, like say their own northbridge chip for tying the CPU to RAM and such as they currently do when the whole package is sitting there waiting for them to turn into something pretty, and by using it they save all the R&D that they are currently putting in. Intel is also really into offering hardware design and integration subsidies, for when you're running into trouble trying to put together your solution, and there are times when it's just a god-send to have one company you call.

Apple is moving to truly commoditized hardware here, not just some standardized parts like PCI or RAM, and while they can charge a premium for their brand, its design, and OS X, since the underlying hardware is the same they can't go go charging crazy margins as only a very select few will be down for buying a diamond-studded iMac. They'll probably use as complete a solution as they can, not hobble themselves but joining the majority hardware party and then making themselves making extra work for themselves.

Compared to say, the Opteron, there is going to be a period where their machines are unexciting, but they'll still have the PowerPC for awhile, and by committing to all-Intel, all-the-time, they're on a much more competitive playing field for the future. Remember, Apple has not been rolling in profit, especially their computer division, and transitions aren't easy or fun, and they want all the slack they can get.

And of course, this isn't a two-year deal for Apple, and it doesn't want to go through a bunch of crap again. Right now, the Xeon has some major problems when it comes to competing against the Opteron. 2004 was a really rough year for Intel in the performance area, and they've had a couple of bum bets, like Rambus, which was their long-term tech solution to the bandwidth problem, only it turned out that Rambus -- as a company -- was pretty much just insane and started suing everyone they could think of.

They were beat to the 1 GHz punch, and perhaps... overcorrected... with some of the design decisions of the Pentium4, which were made all the worse by having to hobble the design to get it out the door at a 'reasonable' time and a 'reasonable' price.

However, if it seems like there is a slight tinge of... desperation... to some of AMD's antics, it's because it gets what position its really in. It's a cruiser that's outmaneuvered a whole damn fleet, and while it's far enough away to start yelling taunts, the fleet is turning around and that gap is going to shorten. It can't do it overnight, but when it opens up, if it chooses to really open up, AMD is going to be in serious trouble, which is why it's trying to build as much hype and mindshare as it can right now.

The stupidest thing someone could do right now would be to discount Intel, or to think in terms of one year as opposed to five. They're one of the largest companies and most resourceful companies around, and the talent they have on-hand is extraordinary, and like Microsoft, they've shown they can get sidetracked down wrong paths from time to time, but when they really set their sights on a clear goal, it generally ends up happening.

note: Yeah OK, it looks like I probably could have gotten away with calling this "Deconstructing AMD", but then I'd have to add chapter headings and such and, you know, look stuff up. Bygones.

yummy alcohol posted button Posted by drunkenbatman
    July 15, 2005, at 10:41 PM


Comments (68)




Post a comment



Anonymous comments are allowed, but please enter something for a name.

And do endeavor to appear sane.









Remember personal info?