Cheap Chips

I’ve mentioned this before, but it bears repeating.

Over the next year, AMD will be shuffling around where it produces and what it produces.

Austin is getting out of the AMD CPU business; UMC is getting into it.

AMD will go from making Athlons and Durons to Hammers and Athlons.

This is how they’ll do it:

AMD is beginning to make Thoroughbreds in Dresden. Preparations are being made by UMC to produce Bartons (i.e., Thoroughbreds with a 512K cache). Austin starts phaseout of Durons.

As Durons are phased out towards the end of 2002, lower-speed Dresden Athlons will be substituted for them. Rather than falling off the map at about the $90 point, they’ll just keep going. In all likelihood (but see below), down to the $40-50 level like Durons are now.

Dresden will start making Hammers and Austin will stop making Durons the last quarter of 2002, but most production at Dresden will continue to be Athlons. It’s unclear how long Palominos will be around. On the one hand, low-speed Throughbreds will be cheaper to make. On the other hand, you have .13 micron refitting expenses that will have to be recaptured within about a year. I suspect AMD would prefer to go .13, but will play it by ear, and if business isn’t too good, they can live with the Palominos.

Dresden will be finishing up its build-up in 2002, so in all likelihood, Hammer production will be coming out of new capacity.

UMC may make some Bartons at the end of 2002, but probably won’t get serious about it until the beginning of 2003.

At that point, AMD will see what happens.

Dresden will not make Bartons; UMC will not make Thoroughbreds. If Hammer takes off faster than planned, Thoroughbred production at Dresden will be reduced and Barton production at UMC will be increased more quickly, too. If necessary, lower-speed Bartons can easily be introduced.

If not, UMC will get less business, at least until AMD starts .09 micron migration. AMD would like to make Dresden a Hammer-only operation as quickly as possible and shift socket A CPU production to UMC, but UMC is a valve that can be opened up more or less as the market demands.

At first sight, it looks confusing, and I suspect the primary reason why AMD hasn’t been too aggressive lately. They’re trying to get their house in order for the next few years.

However, once they get over this hump, this arrangement gives AMD much more production flexibility than they’ve had in the past. Dresden is meant for Hammers, but can make the other stuff if it has to. UMC can make as much or as little of the cheap stuff as needed. That’s a lot better than “make a Duron or make nothing” at Austin or “make an Athlon or nothing” at Dresden.

The UMC chips may cost AMD a bit more, but as we’ll see, this is a minor consideration given what the enemy is doing.

There Are Retreads, And Then There Are Retreads

“Hmmmm, Willy didn’t do too well against AMD last go-round because he was half-crippled. I got it! Why don’t we cripple him MORE and send him out to fight the same guys again?”

Amazingly, this is precisely what Intel is doing with the Celeron-128.

Of course, they have their reasons for this, but they aren’t good ones.

It takes time for Intel to convert production over to .13 micron. They have plenty of .18 micron production waiting to be converted. They have to do something with it in the meantime.

The greatest flaw of the original Willamette was its die size. It was much bigger than the AMD chips, and cost a lot more to make. Not exactly a prime candidate for your cheapo chip.

I bet the biggest reason why Intel cut the cache to 128Kb was to somehow cut manufacturing costs (though there’s no smoking gun in the datasheets to prove that).

Even if they did, a Willamette would still be considerably bigger than a Palomino, and if they didn’t, we’re back to the same old same old.

For newer processors, Intel plans to mostly offset the increased costs of bigger die size by using 12-inch rather than 8-inch wafers, but that’s not a consideration here.

Up to now, Intel’s increased cost hasn’t mattered all that much AMD claims notwithstanding since they’ve been able to get roughly about a third more money for their CPU than AMD has been able to get for its.

That’s not going to be the case in this matchup. Unless Intel is ready to flat-out lose money on these chips (which they might, see below), this is the arena where AMD is quite correct in its cost claims.

So Intel is throwing out a high-cost cheapy chip against a much lower-cost cheapy chip.

Even Intel knows this isn’t the greatest idea in the world; it looks like they want the .18 micron Celeron to be short-lived, too. They’ll go .13 micron as soon as they can, but the bread and butter Northwoods have to come first.

Throw A Punch At AMD And Hit Via, Too?

When you have lemons, make lemonade. Intel showed last year it was incredibly good at that with the Willy/845 SDRAM package. Successfully selling that must have emboldened them.

Remember that Intel doesn’t just make CPUs. They also make motherboard chipsets, and their main competitor is this company named Via, which grabbed a huge chunk of chipset marketshare when Intel was having RDRAM regurgitations.

The Intel chipset folks probably look at Via like Palestinians look at Israel, and over the next few months, we might see their version of a suicide bombing.

If Intel looks upon their Celeron-128 and their new 845-GL integrated chipset as a sort of one-two punch, things could get rather interesting.

As any boxer knows, it’s best to coordinate both hands. It’s a lot better than having two boxers with only one hand, which basically is what AMD and Via are.

This is also OEM land. The battleground here will be the boxes Joes Sixpack and Suit will be buying.

Expect Intel to be rather . . . uh . . . flexible in its OEM CPU and chipset pricing the next few months.

Actually, expect it to look more like tag-team wrestling than boxing, with Intel’s partner being . . . uhhh . . . Dell. Expect to see some remarkably sweetheart deals from them. If the other OEMs want the same thing, well, be like Dell, and just sell Intel. At the least, that will force the OEMs to demand lower prices from people like . . . oh, AMD and Via.

Heads I win, tails you lose. Using Dell as the tail that wags the dog is an awfully good tactic, and just having a much better product like AMD means little in this world.

Of course, you’re not the CompUSA Crew or Disciple of Dell, just occasionally a consultant for a few of those.

We know prices on low-end XPs are already pretty low. We know they’re going to get lower. We either know (or will know shortly) which motherboards will support Thoroughbred. Hopefully, by the time Christmas rolls around, we’ll have a clearer idea what will support Barton.

But even assuming the worst, going with a dirt-cheap 1600+ or 1700+ with a minimal upgrade path to a future dirt-cheap 2600+, a pretty likely 3000+ and possibly even more isn’t the worst plan I’ve ever heard for the average person. Probably not bad for many if not most of you, either.

Assuming just a 2600+, that’s as fast or faster than 99% of the machines owned by the people reading this today, much less the general population. If you handed Grandma a 2.53GHz PIV machine right now for email and browsing, just how long do you think it would be before Grandma told you, “This machine sucks! Get me a new one?”

The Beginnings of Bifurcation

The example of Grandma points to a much bigger issue.

In the past, I’ve spoken about a near future where the computer market would split. There would be computers for a relatively tiny minority who will always need (or at least want) more, and everybody else.

This is beginning to happen.

“The customers seem to like our high-performance stuff. We are selling all the model 2100+ and all the Model 2000+ and then we’re selling the low end stuff and the Durons. There doesn’t seem to be any middle market. It’s probably because the value seems to be bi-modal. If you want performance, you can get great performance using an AMD Athlon and if you just want a low-end you can get a Duron, which is better than a Celeron.” — Jerry Sanders, AMD 2002 1Q Conference Call

“Thinking about buying a new PC . . .? . . . Spend a lot or spend a little. High-end boxes will run you upward of $2,000; cheapies go for $750 or so. . . . There’s a choice [at] $1,500, but I’m betting that you don’t want that. No one else seems to. (I stumbled across this bimodal price distribution, by the way, after examining nine months of data and user traffic from CNET.com). . . . A staggering 30% of all CNET home desktop leads
went to machines that cost less than $750 . . . with a third of those costing $500 or less. . . . And these are perfectly capable [1GHz+, 128-256Mb, 20-40Gb] machines.–Steve Fox, Computer Shopper, June 2002 (sorry, no Web link, just print).

My God, when even Computer Shopper has articles essentially saying “what the hell do you need all this for?” you know the times are a-changing.

This is what AMD is prepping itself for. This is what born-again Christian Chen Wen-Chi probably has been praying to Jesus for every night since buying Cyrix, and it’s beginning to look like Jesus is hearing him.

On the one hand, this is great news for the Joes. They’ll get dirt cheap crap, but remember that crap is good for vegetables (which computationally is all they are, and all they need). On the other hand, it may not be so good for the other end, especially for gamers. If people will pay close to anything for a high-end processor, that’s just what they’ll pay. If not enough people are willing to pay a lot or at least prices inbetween a whole lot and a pittance, you may well see fewer pay even more.

No doubt unintentionally, AMD looks like its actions will force the issue first. Early in 2003, it’s going to have really cheap XPs and comparatively really expensive Hammers. Will the first kill the sales of the second, even after prices slide a bit? Especially after prices slide a bit?

Ask yourself, if you get used to $50-75 for a perfectly decent processor a year from now, what will it take you to spend $400, or $300, or even $200 again? More importantly, what will it take your sister, or jock brother, or Dad or Mom or partner, etc., etc….? You don’t have to tell me, just ponder that.

That’s probably the biggest issue facing Intel and AMD execs right now, not anything else. This is the first time when people can say, “Why do we need more?” and there’s not only no good answer now for the average adult buyer, but no likely suspect visible down the road, either.

Ed