There’s a lot of information floating around about what Intel and AMD are going to do over the next year or so. This will
hopefully tell you what you need to know. Part One talks about the general environment over the next nine months. Part Two talks about the CPUs and platforms we’ll see. Part Three talks about overclocking aspects.
First, though, the general environment:
Willamette–The Operation Succeeded, But The Patient Died
Willamette is likely to become Intel’s next disaster. It’s the wrong product at the wrong time aimed at the wrong people. It’s important to understand why.
Willamette’s design represents two bad underlying attitudes Intel has: “We’re cheap” and “We’ll always be the center of the universe.”
Back in the mid-nineties, Intel figured that a processor design had an effective life of two years. They planned on two design teams alternating generations of processors.
So the Pentium Pro came out in 1995, and Willamette was supposed to debut in 1997.
Didn’t quite work out that way.
Design time went into the Merced/Itanium boondoggle, and the Pentium Pro core was extended and extended and extended. This gave AMD a chance to catch up. Willamette was put on the back burner until AMD was well into Athlon development, and Intel belatedly
realized they now had some competition.
The core Willamette design is meant to last for a while, and that’s the problem.
Intel made a number of design decisions that no doubt will make a lot of sense a year or two from now.
But Intel doesn’t need a really good product a year or two for now, they need one now, and the Willamette coming out in a month or so isn’t. Even Intel’s conceded that by making the .18 micron Willamette platform a nine-month wonder until the “real” .13 micron Willamette and platform shows up.
Willamette really isn’t meant to be a .18 micron product. The die size is too big to make economically. There’s not a ton of room for rampup at .18 micron when you’re starting at 1.4 Ghz.
More importantly, Intel decided to change its design to have its CPU do less per clock cycle more quickly. That will make sense down the road when the chip is running at 2 or 3 Ghz. For now, though, that just means
it won’t compare too well clock-for-clock with its own current product, and especially against the AMD competition.
Provided AMD’s core revisions in a couple months don’t yield the same results, the initial Willamette should trail AMD’s products clock for clock by a considerable margin in most standard benchmarks.
For a little while, the overall numbers will look better when a 1.5 Ghz Willamette is being compared to a 1.1 or 1.2 Ghz Athlon, but that’s not going to last long.
PR and Vaporware
Intel’s strategy seems to be the following: Keep getting Mhz media hits with Willamette while not making much product until .13 micron migration. Make up new benchmarks which make your product look better. Since most people don’t buy the very high processors, anyway, with some tweaks to improve yield at higher speeds, current Coppermine speeds can hold the fort until the .13 micron cavalry arrives next June or so. After all, that’s pretty much what they’ve been doing the last year, and it’s worked so far.
Can it continue? Intel will essentially be uncompetitive at the high end for over six months. While it’s true that they’ll keep about 70% of the CPU marketplace simply because AMD can’t supply more than 30%; that’s not going to look so good come the day AMD can supply more than that.
Will the Sheep Start Looking Around for A New Shepherd?
So long as the vast majority of people keep seeing Intel as THE CPU company, this strategy will work. However, the longer Intel comes up with songs-and-dances and slogans rather than products, more and more people are going to realize that the Emperor is really doing a striptease.
Will the media consider Willamette a wonderful product and obligingly not notice that it isn’t quite as good or quite as available as the AMD products? Maybe they will. What if they don’t?
How much damage do you think a few months of “Intel has fallen behind” would do to the image of invulnerability? How much do you think that would change the average person’s impression, and eventually buying decisions?
Remember back in the late eighties, when IBM lost the personal computer market? I remember the computer magazines bending over backwards to say nice things about the IBM machines, until they just couldn’t do it and retain a shred of credibility anymore.
Not saying the mainstream computer media will turn on Intel in the next year, but expect to see some start hedging their bets.
A company like Intel doesn’t fall in a day or a year. They can bank on their reputation for quite a while, just as IBM was able to in the mid-to-later eighties. IBM kept marketshare pretty well with overpriced but good machines. The crunch came when people realized other machines were not only cheaper, but better.
AMD’s Sneaky Two-Step Revolution.
AMD is a sneaky company. They announced sweeping price cuts, and the conventional wisdom said it was the usual AMD move of desperation.
Before, AMD could only do a one-step, cheaper. You can only go so far with that before you hurt yourself and make your audience think it isn’t worth much, either.
That’s not the situation here. Thanks to Intel’s decision to essentially do nothing besides put up Willamette Jr. as a strawman for the next nine months, AMD is ready to do a two-step, first cheaper, then better.
Early next year, people are going to see AMD products that not only are cheaper than Intel’s (let’s leave overclocking out of this for the moment) and more available than Intel’s, but clearly better. They’ve never been able to do that before.
A high price only helps you when you sell it. Intel has always had a skewed pricing structure; milking a few while they could. Note that the AMD price cuts essentially abandons this; the price cuts were mostly on the high end, where they probably sold very few processors, anyway. The pricing structure is now less skewed than Intel’s. People are far more likely to spend $500 for top-of-the-line than $1,000. I’m seeing a lot more people saying, “I’m going to buy a 1Ghz Thunderbird” than I’m seeing, “I’m going to buy a 1 Ghz PIII.”
When I go to computer fairs, I’m seeing 1 Ghz TBirds for sale; I’m not seeing 1 Ghz PIIIs.
Remember something about AMD and Intel. Intel has historically had an ASP (average selling price) of roughly $200. AMD’s has historically been under $100, sometimes a lot less than $100. When you’ve survived getting about $50 for most of what you make, having Durons going for $70 looks pretty good. Selling your best product for $450 looks pretty good when you’ve usually gotten only about $200 for it.
So AMD can cut Athlon prices quite a bit, and still do a lot better than they’ve done in the past. For Intel, it’s all downhill. AMD’s high-end price cuts are forcing Intel to cut out a big chunk of pure profit from what had been their most profitable processors.
Now add to that relative improvement of the product:
In 1998, if you had to choose between a K6-2 400 costing a couple hundred dollars less, or a PII 400, what would you have preferred?
In 2000, if you had to choose between a Thunderbird 1Ghz costing a couple hundred dollars less, or a PIII 1 Ghz, which would you have preferred?
In 2001, if the choice is between a Palomino 1.5 Ghz costing a couple hundred dollars less, or a Willamette 1.5 Ghz that trails by about 15% in performance. which would you prefer?
Your opinion shifts, doesn’t it? Even if you stick with Intel, the choice gets harder and harder, doesn’t it? Cheaper and better.
From the surveys we’ve had, there’s already been a tremendous shift in attitude about AMD. Two years ago, the vast majority of you would not have considered an AMD product. Now, the vast majority of you will.
You still may not buy one, but I no longer hear “the chips are inferior” but rather about mobo problems. What happens if and when the mobos get better?
Even for overclockers and even if AMD has essentially prevented more than nominal overclocking for future generations, you may still be a little better off with AMD.
But even if you’re not, the percentage of people who will at least consider AMD will continue to grow, and more of them will buy.
An Early Sledgehammer for The Server Segment
Intel has had a real racket going on with its Xeon processors. Add some more cache, and charge an arm-and-a-leg for it. I’m sure it costs Intel more to make them, but nothing like the price they charge for them.
Well, AMD is really going to have those folks stuck between a rock and a hard place with its Mustang processor. At least on paper, it will look a lot better as a processor than any Xeon. I’m sure AMD will charge a whole lot less than Intel.
I know, reliability counts for more than speed, and the chip and especially the infrastructure surrounding it is completely unproven. I know the average person buying a server is far more conservative and less price-conscious than the average desktop buyer. This certainly won’t be an instant hit for the Fortune 500.
But not only the Fortune 500 buys servers. Some of you have servers just to have Quake matches. There are plenty of people with websites who’d like a better server, but can’t pay Intel prices.
If you thought AMD was in clover making $500 from a processor, what do you think $1,000 each for a line of processors must look like?
So AMD can grossly undercut Intel in this arena, and still make a ton of money.
What’s Intel going to do? If they cut their own prices, that will hurt their bottom line by far, far more than it will hurt AMD. If they don’t, AMD can grab itself market share from the low-end, and if the systems prove reliable, move on to Corporate America.
The Intel Era Is Ending
That doesn’t mean Intel is going away (though it eventually could, see below). What that means is that the environment under which Intel has operated is going away. Intel has spent its existence fat and happy with fat profit margins because they never had any real competition. Those days are ending.
They won’t be able to charge $1,000 for a desktop processor anymore. They won’t be able to charge $2,000 or $3,000 or $5,000 for a server processor fairly shortly.
Most of the computer industry has become a commodity industry with low-profit margins. The CPU segment will be next.
Many of Intel’s recent errors and blunders can be traced to this. To preserve their 60% profit margins while prices are declining, they’ve cut costs to the bone and beyond. They’ve tried to stretch out physical and human resources, and it’s starting to show. They are trying to hold back the tide.
Right now, Intel hasn’t really met profit expectations for over a year. What they’ve been doing to meet and exceed those expectations lately is sell off their stock portfolio, with a nudge-and-wink from the stock analysts. They can’t do that forever.
Should Intel stay in the CPU market, they will inevitably become a less profitable company. Still pretty good; not going to become Craig’s Cheap CPUs, just nowhere near as good as in the glorious past.
In contrast, from a profit perspective, AMD until recently has been in the profit sewer. Anything is an improvement.
Again, Intel has generally had an ASP (average selling price) of roughly $200 per CPU. Until recently, AMD’s was about $70. An ASP of $150 for Intel would be disastrous; it would be paradise for AMD.
The erosion continues. In all likelihood, the result of all the items I’ve mentioned is that AMD’s market share will go from around 20% to around 30%. No catastrophe yet, just a few more beachheads AMD grabs.
Unless AMD really blunders, though, the problem with Intel is that there is no light at the end of this tunnel. Even after Intel moves to .13 micron, that’s hardly going to blow AMD away, they’ll move to .13 micron shortly thereafter. Towards the end of the year, AMD introduces Sledgehammer, which may
be a better 64-bit solution for the average computer user. Right now, Intel wants to make $4,000 or $5,000 out of its 64-bit processors, and never mind the rest of you. Sledgehammer could steal the desktop market away while Intel dawdles. If that happens (and these are big ifs), it might be the processor that convinces Intel to sell its CPU business, downsize, and look for greener pastures.
This is NOT “Good triumphs over evil.” I don’t think for a second AMD would be one bit better than Intel in the driver’s seat. What I do think is possible is that AMD will reduce profits in the industry and become competitive enough to the point where Intel won’t be able to play the old games anymore, and can’t play under the new rules well enough, so they’ll sell out to somebody who wouldn’t mind
making 30 or 40% profit margins.
A year ago, a scenario like this was completely inconceivable. It still looks unlikely now. Two or three years from now, it could be reality.
2001: Your Next Odyssey
If you’re looking for the next major advances, they’re not going to happen this year.
We may see products being announced and introduced this year, but they will be either unavailable or unripe while the calendar says “2000.”
Let’s see what we can look forward to:
Intel: Giving Us the Willies
For the rest of 2000, the only real production introduction we can expect comes from Willamette.
While normally, this would be pretty significant news, it won’t be for the average overclocker for the following reasons:
1. The chips will be expensive. $500-600 to start. Intel won’t be making too many of them, either, so expect that price to remain stickier than usually. Don’t expect a Willy to cost less than $300 until sometime in the spring, by which time, you might as well wait for a .13 micron chip.
2. The infrastructure surrounding it will be on the expensive side.Willy mobos will be costly. However, RDRAM will no longer be the whipping boy for cost.
Provided you buy 256Mb of RAM, RDRAM costs right now are not much higher than DDR prices are going to be.
If you buy Willamette now, you have to buy 2 modules to take advantage of the dual-channel structure of the Tehama motherboard. You can buy two 128Mb RIMMs for about $450.
In contrast, 256Mb of DDR looks like it will cost about $350-375. Sure, DDR will be cheaper, but the price difference is no longer absurd, and Intel’s price subsidies will eliminate most of that difference.
However, if you only buy 128Mb of RAM, the cost of two 64Mb RIMMs is not much less than buying two 128Mb sticks, while buying one DDR stick costs half the price of two DDR sticks.
All I’m saying is that cost is no longer a killer reason not to buy RDRAM.
3. The initial chips won’t perform all that well for their speed. Figure a Willy on the whole will do about 80% of the work of a PIII or Athlon at the same clock speed. So buying a 1.5Ghz Willy is really like buying a 1.2Ghz PIII or Athlon. That’s not a terribly good deal.
This doesn’t mean Willamette is a bad chip. Intel designed the chip with plenty of headroom to be cranked up in later generations, much more so than if they stayed with the older PIII design. Problem is this isn’t a later generation, but the first. Willamette looks to be a better .13 micron chip than it is a .18 micron, just as the PIII is a better .18 micron chip than it was a .25 micron chip.
4. Overclockability A Big If With Big Factors Going Against It
The chip is huge as CPUs go. It generates a lot of heat. By the time it becomes affordable as overclockers define it, something better will be coming shortly thereafter. Just like buying a Katmai a month or two before the Coppermines came out.
Putting all that aside, there’s been some tentative hints that if you give it tender loving care, a Willy will be an OK overclocker, and that an overclocked one will do a bit better than the AMD competition. I wouldn’t be surprised to see a few getting 2Ghz or a little better out of one next spring.
However, it doesn’t look like a good bang-for-the-buck overclocking choice.
PIIIs, the next Celerons?
Intel says it’s going to improve the PIII a little more by adding more cache (up to 512Kb), shrinking it down to .13 micron, and increasing the FSB.
That’s nice, and it would be great if they did it right now, but it looks like the middle of next year. At that point, it looks like too little, too late. You’ll be able to buy an enhanced TBird that will do just about as well long before that, and you’d rather buy a .13 micron Willy over that.
Analysts are hinting that Intel is afraid that an enhanced PIII will look too good in comparison to the Willy available at the time, and thus will handicap it a bit at introduction, probably by making the initial processors run at a 133Mhz bus.
That might prove good for overclockers (see Part III for that), but what is clear is that you’ll need a new mobo and RAM to take advantage of it. If you’re going to spend that kind of money, it would seem you’d be better off with one of the other alternatives.
Unless Intel moves the PIII downscale to fill the spot now covered by Celeron and phases Celeron out. Such a chip would be very good competition against Son-of-Duron. However, that makes sense, so Intel probably won’t do that, but rather try to position the PIII as some sort of midrange chip with pricing higher than Son-of-Duron to maximize revenues. That will not make sense and prove to be another Intel blunder.
For now, cC0, and That Give Me the Willies
Intel’s announcement that we’re not going to see 1.13 PIIIs until the spring bothers me. There’s two plausible reasons for it:
- The problems found in the cC0 chips are harder to fix than they originally thought or
- Intel’s being cheap, and figures they can save money doing the fix when they migrate to .13.
I suspect it’s the latter.
In the meantime, there’s a big question mark in my mind how reliable a cC0 chip is going to be if you crank it up past 1Ghz. I could be right; I could be wrong; I just don’t know for sure.
AMD: Tweaking The Core and Competition
Jerry Sanders announced yesterday that we won’t see the next generation Athlon chips until early next year. Not surprising. If you can wait, and want DDR, this are the chips to wait for. Provided AMD cooperates, I expect this combo to be the hot overclocking system for the first half of next year.
As expected, TBird production shifting heavily to Dresden, expect to see many more blue meanies out there over the next couple months. Durons will remain aluminum and stay in Austin for quite a while.
AMD will announce a 1.2Ghz processor soon, one more speed bump before the end of the year.
Expect to see people start to do a little better overclocking AMD chips the next few months, but not a ton better. Maybe a 10% improvement over the next couple months.
You have three players in the DDR game: AMD, Via, and Ali. The article at Ixbt Labs lists mobos that have been announced.
Expect to see lots of announcments and reviews, but no real product until the tailend of the year.
All you really need to know at this point is:
- If you want to overclock, you don’t want PC1600 memory, you’ll want PC2100, and you’ll want a mobo that can run at 266Mhz FSB.
- The DDR boards for AMD chips will give you a boost; the initial DDR boards for Intel won’t. That’s because the initial Intel boards will still be running at 133Mhz FSB, not 200 or 266 like the Athlon boards. Later next year, we’ll see true DDR boards for Intel chips, just not soon.
Next, the big issues and question marks for overclockers.
A Whole New World
New families of products bring new issues. These are the questions for which you need answers to figure out if something is right for you. If you see previews and reviews over the upcoming months, look for answers to these questions. If they aren’t addressed by the review, ask.
Depending on the answers, we may be in for some pretty good times, or things could get rough.
No matter what, upgrading to the next generation is going to cost you. We’re talking about CPU, mobo AND RAM for most people. For most people, we’re looking at DDR, for a few RDRAM.
Should we keep the pencil sharpened?
Issue one is whether or not AMD will continue to allow changing the multiplier. The processors that will be debuted early next year represent a major core revision, and that gives AMD the opportunity to make those kinds of changes.
Let’s face it, AMD didn’t exactly rival Fort Knox when it came to locking its processors. They went through the exercise in the first place to make somebody happy, and that person could hardly be thrilled if he or she has found out how “tough” that lock is to break.
They could remove the L1 bridges. They could remove the pins that make it possible for the mobo to tell the CPU what to do. They could do nothing. But this new generation brings back another possible way to overclock . . . .
The biggest questions about the next generation of AMD processor revolve around these two numbers.
Up to now, Athlon motherboards have had one official speed: 100Mhz (in AMDspeak, 200Mhz DDR). Most of the Athlon DDR motherboards coming out with have two official speeds: 200Mhz DDR and 266Mhz DDR.
The big FSB question is:
- Will the Palomino chips come in 200Mhz and 266Mhz varieties, like Intel chips come in E and EB, or not?
The smaller questions are:
- Can I overclock old or new 200Mhz Athlons to 266Mhz?
- How far beyond 266Mhz will these motherboards go?
AMD has a pretty tough decision to make on the first question. If they want to sell more processors, they’ll provide a 200Mhz version so that people with current socket A can upgrade. Another reason why they might want
to do that is the question of how available PC2100 RAM is going to be. I have the feeling that making PC2100 is a good deal more difficult than the DDR people are letting on, and that PC1600 may be the DDR memory most available at least at the beginning.
If AMD does come up with a 200Mhz version, then we’re in clover. Buy a 1.1Ghz Palomino, push the FSB up, bypass whatever annoyances AMD comes up with, and you’re running at about 1.5Ghz.
On the other hand, if they make these new processors 266Mhz only, it’s going to foreclose more than nominal FSB overclocking. That will leave only the multiplier change.
If AMD really wants to shut the door on most overclockers, the time to do it is with these new chips. If they eliminate the L1 bridges or pins and make only a 266Mhz Palomino, they’re serious. If they don’t do both, they’re not.
What will they do? I really don’t know.
Will We Need A New B21 Pin Trick?
The next question is really asking “Will there be blocks that will keep a 200Mhz processor from running at 266Mhz?” The answer to that will probably be “Yes,” but I doubt that will be much of a problem. Since just about all of the announced mobos are designed to run at either speed, a workaround should be possible.
Or a mobo manufacturer will simply not implement any block.
How far up the FSB ladder?
I suspect not all that far. From first signs, the next step beyond PC2100; PC2600, is proving to be tough. On top of that, Via already has chipsets that can handle 155Mhz. I suspect they’ll save themselves some design time, and just implement a DDR version of that, which would give you around 300-310Mhz.
We know a good deal less about Intel’s offerings than we do AMD’s, which is understandable because
1) the product they’ll be introducing shortly is really a new product, and we need some hands-on experience with a new set of equipment
2) everything else won’t be seen for another six-nine months.
Therefore, my comments are going to be a lot vaguer about Intel. We don’t know enough to really formulate specific, make-or-break questions, and I suspect some issues are still up in the air.
Nobody Knows Diddly-Squat About Overclocking Willy
We don’t know the CPU, we don’t know the chipset, and we don’t have much O/Cing experience with dual-channel RAMBUS. Three strikes and we’re out when it comes to talking about this now. When we know something, we’ll tell you.
Does Intel have a Cu-T for you?
About a month ago, Bert McComas talked about a new Coppermine stepping called the Coppermine-T. He didn’t have too many details other than it would work at 1.13Ghz and could run at either 1.2 or 1.5V(?) and could run on existing Coppermine platforms. It’s not clear if
this got cancelled by Intel’s postponement of a new 1.13Ghz chip until sometime in 2001.
Update 10/16/00: If Tom’s Hardware is right, we will see .13 Coppermine-Ts that should work in “current equipment” next spring/summer.
I put the term “current equipment” in quotes because unless Intel has a wonderful surprise for us, these Coppermine-Ts are all going to be 133Mhz CPUs. That doesn’t give you much of an overclock for two reasons.
1) Memory: Unless SDRAM gets better in the next couple months; it’s going to be rough even with the best memory to do more than 167Mhz, and that certainly is no guarantee.
2) PCI bus speeds: At 167Mhz, with a PCI divisor of /4, you’re talking about a 41.67Mhz PCI speed, same as 124Mhz with a divisor of /3, and 83Mhz with a divisor of /2. While some equipment can handle that, I would say a majority would not.
If you already have an 815 board, high quality RAM, and can run right now at 160Mhz or more, Coppermine-T might be a nice upgrade chip. If you can live with little to no overclocking, Coppermine-T might be a nice upgrade chip. If neither fits your bill, an AMD DDR system earlier or a .13 micron Willamette later look like better bets.
Tualatin: All dressed up and no place to go?
.13 micron process. Voltage below the minimal range of current mobos, so you’ll need a new one. More ondie cache.
I’ve previously expressed doubt about this chip as being too little, too late against the AMD competition.
Since it won’t work in current equipment, it’s not an upgrade option. That’s really a shame, because it would make a really good one.
Finally, its overclocking potential is unclear. Not the chip itself, just the .13 micron shrink should give maximum speeds a big boost.
It’s the FSB that creates the doubts. The best guess is this starts at 133Mhz, then goes to 200Mhz. The problem is: 200Mhz what? If it’s 200Mhz DDR, that raises havoc with any multiplier lock. If it’s 200Mhz SDR, there isn’t 200Mhz SDRAM around.
I suspect you’ll be able to overclock decently one way or the other, but I’m not sure it’ll really matter.
Unless they make this very cheap, it’s hard to see why you would buy this. AMD will have equivalently-powered equipment months before it. If you want more power, buy a .13 micron Willamette.
The only way this makes some sort of sense is if Intel plans to keep Willy expensive, and Tualatin’s maximum speed is around 2Ghz. If both of these are the case, then Tualatin might become an overclocking favorite the second half of the year. AMD will probably slow down to migrate to .13 micron itself, and get ready for Sledgehammer.
Don’t They Like Willy?
Intel is muttering things about not expecting Willamette to become the bread-and-butter processor for Intel until 2002, so I guess they think Tualatin is going to do the heavy lifting for the second half of the year. That’s a lot to ask from the ancient Pentium Pro design.
I can see why Intel doesn’t want to go into mass production of an .18 micron Willamette. I think they’re wrong, but I can see it. But why won’t they ramp up a .13 micron Willy quickly? Why waste precious fabricating resources making Tualatins when you have a better product ready?
I can see Tualatin as an insurance policy. I can see Coppermine-T as the same. I can see keeping things like Almador as insurance just in case Willy falls flat.
But if Willy’s OK at .13, why dawdle?