Some comments on a few websites have been revised, and now the FX-55 is being reported to make its arrival in mid-late third quarter.
This estimate makes sense given what else AMD has been saying about what it’s doing.
By that time, AMD ought to have at least the first fruits of its 90nm production, and that’s what the FX will almost certainly be.
The FX-55 ought to provide an early demonstration of why we have for the longest time suggested that people wait for 90nm Hammers before leaving their socket A platforms.
First, unlike current FXs, the FX-55 will be at the low-end of the 90nm totem pole rather than at the high-end of the 130nm pole. That should make it inherently more overclockable than its ancestor.
Even a pessimistic estimate of what the initial ones ought to do would get an FX-55 close to 3GHz, and the Mr. Freezes should certainly be able to break 3GHz and then some. After the second spin of the production silicon, 3GHz will probably be a reasonable little-to-no-fuss effort.
3GHz is certainly nothing to sneeze at when you realize that conservatively, that translates out to better than a 4.5-4.8GHz Intel Prescott. Intel won’t reach those kinds of speeds until they get to Tejas.
A 2.6GHz FX-55 ought to run rather cooler than an FX-53, and downright cold compared to a PressHot. There’s some rumored indications that 90nm SOI chips run very coolly at low speeds, but ramp up quickly at higher ones. Even at high speeds, though, power and heat are no big deal by today’s standards and certainly much better than an equivalent PressHot would be.
The die size of a 90nm Hammer ought to hover around 100 sq.mm, a bit more with a big cache, a bit less with a little cache. Per number of chips on a die, it will be like making XPs.
This ought to make AMD try to shift Hammer production over to 90nm for all its speeds as quickly as they can.
Paris: Tactics Over Strategy
Obviously, not too many of you will be standing on line or online just waiting to fork over $750 for this chip. AMD looks to have two other options for you.
For the relatively loose-of-wallet, there will be 90nm Newcastle socket 939 processors. They’ll have less cache, and probably a lower-rated speed than the FX.
For the low end, we’ll see “Paris,” which essentially will be a Thoroughbred B with an online memory controller but no x86-64 and just a single-channel memory architecture.
Paris is a rather odd duck. It has a rather big conceptual problem is that it basically gives the finger to any claims that x86-64 as the wave of the future. The more boxes that have x86-64, the more likely it is that mainstream software writers will write for it, which will put pressure on Intel to go universal with it, too.
Yes, AMD does get an advantage from doing this in that people interested in x86-64 will be forced to move upstream and pay more to get it. It also provides a definitive difference in performance between Paris and its big brother.
But these are tactical advantages that can blow up in AMD’s face the moment Intel ever decides to put out an x86-64 Celeron. Even if that doesn’t happen, you know there’s going to be people with Paris OEM boxes who’ll think they have x86-64 capable boxes, and when they find out they don’t . . . .
This is a great example of AMD thinking tactically rather than strategically. If they really want x86-64 to become a mainstream standard rather than blue crystals for geeks, they’d include it. True, they may lose some low-end Newcastle business if they do, but they’ll probably lose even more Newcastle sales due to lack of mainstream x86-64 software to Intel if they don’t.
That being said, the blue-crystalless Paris has rather differing levels of appeal to different groups.
It looks great as a cheap Joe Sixpack box, extending its lead over any Celeron box to almost ludicrous proportions.
It doesn’t look as hot as an Athlon XP replacement. The chip will probably be priced right for AMDers, but add the cost of a mobo to it, and we’re probably looking at $200 for a replacement. The price/performance improvement will be less than compelling, at least to start.
It won’t start looking good until we see 90nm equivalents in 2005.
What will overclockers do? Will they go for 90nm Newcastles sooner, or 90nm XP+s later?
It all comes down to price.
The Middle Is What Matters
The real key to whether AMD will succeed or not is how they position the socket 939 Newcastles in price range. The entry-level Newcastles ought to be priced roughly in line with the prices of Intel mainstream processors, and the sweet spot ought to be around Intel’s, which is around $170.
The danger is that AMD is going to considerably overprice these chips, and keep not only most overclockers, but also most mainstream OEM buyers from buying them.
It’s stupid to give people a choice between a $300 or even $200 chip and a $90 chip when most people are walking around with a big price bias in their heads. They’ll pick the $90 chip almost every time. AMD overclocking fans will do just that if that’s the choice you give them.
No, what you do is give them a choice between a $140 and a $90 chip, and make the $140 chip look like a value meal. That leaves the consumer feeling smarter and more upscale, and gives you fifty more dollars for something that might have cost you five dollars more to make.
It works for Intel. It works for the video card makers.
Why shouldn’t it work for AMD?
If AMD is really interested in revenue maximization, what they ought to do is move the low-end up. Make the cheapest Paris $70, and let Intel for once feed the bottom-feeders if so inclined. Make the product range for Paris, say $70-110, and start Newcastle off at $140-150.
That’s how you get people to revalue AMD in their own minds. If people will take a much less capable Celeron to save $20, let them. If people won’t pay $150 for a CPU, then at least get $70 rather than $50 for it.
The way you get your ASP up to sustainably profitable levels is not to get an extra $200 from 2% of the people, or $500 from .5%. It’s to get an extra $20 from everybody, and an extra $50 or $75 often.
It’s not as dramatic, but it gets you more money from your current customer base while leaving you free to shake the corporate tree.
For us, it’s going to determine who is going to overclocking what starting six months from now. If AMD gets a $140-$150 Newcastle out, most of us are going to pile into that. If they price it much higher than that, people will wait until Paris and especially 90nm versions come out and play with that instead, and we go back to being an anchor on AMD’s ASP all over again.
Stretch the pricing parameters. Don’t snap them. Don’t sell gourmet burgers; sell happy meals.
AMD can charge $750 or for that matter $7500 for an FX; it doesn’t make any real difference to the company’s overall finances because only a relative handful of people will buy them.
It’s what the average AMD buyer pays that matters, and sometimes you have to charge less to make more.
Ask any McDonald’s.