AMD's 2004 Strategy

Add Your Comments

As we mentioned the other day, AMD appears to have come out with a new placement strategy for its processors.

It’s not a bad one, provided you leave the soon-to-be orphans alone.

Essentially, AMD will make socket 939 their mainstream and better solution, and socket 754 their value solution. While it still looks like one socket too many, putting the two sockets into very different markets is acceptable and should reduce the confusion down to reasonable levels.

There will be two kinds of socket 939s. There will be the “big” FX type model, and there will be Newcastles for the rest of us, a split similiar to Intel’s Extreme Edition versus regular PIVs.

AMD ought to have an easier time making a lot of chips with 512K cache than 1Mb cache and/or sell “big” FXs where only half the cache works right, so prices shouldn’t be too bad for those.

The same split will occur through the transition to 90nm. You’ll have a “big” and a “little” socket 939 chip.

Some seem to think this quite confusing, but it really ought not to be. The expensive chips with 1Mb cache will remain FXs, and the Newcastles will 512K will be the mainstream chip at more reasonable prices.

Distinguishing The Two

A problem AMD faces is that Newcastles are likely to come pretty close in performance to FXs, presuming that the only difference between the two is cache. Extra cache helps, but not all that much.

Will it be enough to justify a much higher price tag?

That depends on what AMD expects from its high-end chip in terms of sales. If they treat what will be the future iterations of the FX like Intel looks upon Extreme Edition, and expect to sell only a relative handful who buy unconscious of price/performance, and will spend hundreds more for an extra handful of percent improvement, no problem. I won’t buy it, you won’t buy it, but that’s OK, it’s not meant for us.

There will only be a problem if AMD decides that the difference in cache isn’t enough of a difference and tries to cripple Newcastle some more. For instance, if socket 939 Newcastles could only handle one memory channel, that would be very bad, as someone else has already noted.

A few of you might say, “but haven’t you said in the past that there should be less choice, Ed?”

Indeed, I did, but what I suggested was one socket, two mobo types (single/dual channel).

At best (presuming that cache is the only difference between Newcastle and FX), AMD is offering two sockets and two mobo types. If AMD somehow cripples a socket 939 Newcastle, you’d end up with two sockets and three mobo types.

The first is acceptable, the second isn’t.

Which will AMD do? You can make a half-decent argument either way. The first would be the smart thing to do. On the other hand, this is a company that didn’t at least publicly think for months ago that it even needed dual-channel memory desktop systems.

P.S. If AMD does do the dumb thingand makes single-channel only socket 939 chips, might be interesting to see if people watching at home can “undo” it.

Widowed With Children…

Widowed With Children

Whichever way AMD goes with socket 939, the eventual fate of socket 754 is clear: the bargain basement. However, like a chicken with its head freshly cut off, socket 754 doesn’t know that it’s dead, yet.

This is a rather dramatic change of events given that socket 754 was supposed to be the only desktop socket for AMD just a few months ago. This was a big blunder on AMD’s part given a) what Intel planned and then did with Springdale/Canterwood and b) the fact that AMD already had a dual-memory architecture all set to go with Opteron.

Yes, a blunder. Right now, dual-channel is the difference between beating a “regular” PIV and losing to one. In its initial decision, AMD effectively chose to lose, a decision it had to reverse at the last moment and make the Opteron That Wasn’t An Opteron.

All these moves are probably AMD’s way of getting out of that earlier blunder while (probably) not stepping on certain corporate exec toes too hard.

However, the current roadmap has enough wiggle room (especially without any pricing) for this not to be the only interpretation of the data. For instance, AMD might let the two versions of Newcastle fight it out, see which one does better, then revise their plans accordingly.

Either way, though, we all have a vote in our wallets, and whether it’s a done deal or a real contest, it is in our long-term best interest to vote for socket 939.

You might say, “Well, Intel has all sorts of choices, too, from Celeron 100MHz chips to 133MHz PIVs to HT 200MHz PIVS to EE.” Sure, but what does this audience pay attention to? With very few exceptions, they’re just looking at door number 3 because that’s the best combo of performance and price.

We’re going to have to (and will) make the same sort of distinction with AMD, too. The problem we have is that the bulk of those (relatively few) who have bought desktop Hammers so far have bought socket 754 chips simply due to the extraordinary cost of the FX. In the months to come, that will have to be reversed.

Those folks look like they’re going to be orphaned. It looks the same way for anybody buying a socket 754 Newcastle. The action and upgrades will be happening in socket 939, and will only reach socket 754 much, much later.

Now there are those who don’t mind second-class citizenship at the right price. People overclock Durons. That’s fine and dandy. But those inclined to overclock Durons do so because Durons cost $30. Will they do it if they cost $200 or more?

Then again, if you looked at XP overclocking the last year, there was a lot of sensitivity to price even when it was minor. A hefty proportion wouldn’t lay out $90 for an “A” TBredB or Barton when they could lay out $55 for something a little less capable.

Will there be the same level of price sensitivity when it is, say, $500 for a dual-channel vs. $400 for a single-channel system?

Our advice to this audience (and this is leaning rather more towards performance than price), would be to forget about socket 754 and focus on socket 939. You’ll get the benefits of dual-channel memory for gaming now, and 90nm processors rather sooner than with socket 754 later on.

If the prices of the CPUs are relatively close, spend the extra money. If AMD prices socket 939 Newcastle processors far higher than socket 754 processors, sit on your wallet until they change their minds.

Again, if AMD wants to have a version of an extreme edition and charge a ton for it, let them. So long as Newcastle are reasonably priced, that’s reasonable.

However, if they treat all socket 939 processors like it were an extreme edition, don’t let them force you into the cheapy stuff.

Leave a Reply

Your email address will not be published. Required fields are marked *