The Case Against CPUs

Both myself and Jack Russell have written about AMD charging too much for its own good.

Some seem to have misunderstood what we’re saying, and they think we’re saying that AMD shouldn’t charge $500 for a 3200+ next week.

That’s not what I’m saying, and while I can’t speak for Mr. Russell, it doesn’t appear to be what he’s saying, either.

Perhaps it would make matters clearer if we looked at Prescott and desktop Hammers together, at the same time, and explained the problem we see with both.

Objection One

From an overclocker perspective, given the excellent overclockability of current Athlon XP and PIV platform, desktop Hammers at initial pricing offer poor bang for the buck.

You might say, “So will Prescott when it first comes out.” That’s true, we think exactly the same about it, but there’s a few differences between the two.

Objection Two

We will pay little notice to the initial Prescott as a processor a member of our audience ought to buy (as opposed to the general heat issue), for exactly the same reason as we listed in Objection One.

This will remain the case until Intel releases 2800 and 3000 Prescotts early next year (probably mid-January). They will cost $218 and $278 at introduction.

Let us apply the same principles we applied to Intel to AMD.

It is at least unclear whether or not AMD will release at least “big” Hammer at speeds less than 3200+. As of now, it looks like there will only be three models in the first generation for either Athlon FX and Athlon 64. It does not appear that these CPUs will become affordable to at least a sizable fraction of the computer buying audience (“affordable” being rather generously defined as $250).

We think it is unwise for the general success of the Hammer platform for AMD to have so few models of “big” Hammer at such high prices. We think it would be much wiser for AMD to release at least lower-speed Athlon 64s and FXs at somewhat lower prices.

One might say that this is special pleading, that this is just a way for overclockers to save money. This isn’t true for two reasons.

First, AMD’s high prices will stunt all Hammer purchases, which will make it harder for software manufacturers to port mainstream applications to x86-64. Second, if AMD doesn’t provide lower-cost options, overclockers will not end up paying more; they’ll end up saving a lot more money by not buying it at all.

One may say, “Well, perhaps AMD will put out lower-speed FXs/64s later on, or come out with cheap “little” Hammers later on.”

The latter is more likely, but the small tidbits we’ve gotten so far indicate that AMD intends to charge Intel mainstream-like prices for what will essentially be AMD Celerons. With a quarter the cache of the “big” FXs/64s, and a single-channel memory architecture, performance of such CPUs should be significantly less than that of 64d and sizanle less than that of FXs.

But there is an even more fundamental problem than that.

Objections Three and Four…

Objection Three

Even if Intel conquers their current heat problems (and the proper attitude for that at the moment is “Show me”), we are unlikely to recommend Prescott as anything but a CPU upgrade for those already owning current PIV systems or those who replace systems rarely. That is simply because less than six months after that, Intel will shift to socket T, and that will also bring along DDR2 and PCIExpress. A number of months after that, Intel will introduce Tejas.

We currently believe the best bet on the Intel side is to stick with or go to a Northwood system (at this point, at least with a updated mobo that will support for sure Prescott wattages), then wait until at least Tejas for the next major upgrade. We view Prescott as at most an optional CPU upgrade

Let us apply the same principles we applied to Intel to AMD.

One may say, “Well, perhaps AMD will put out lower-speed FXs/64s later on, or come out with cheap “little” Hammers later on.”

This may be so, but then we run into the same instant obsolescence problems we saw with Intel.

We have socket issues. AMD will come out with socket 939 CPUs and mobos. What does that do to socket 940 buyers?

There’s been some rumors that AMD will go solely dual-channel in future generations of desktop Hammers. What does that do to socket 754 owners?

The killer problem, though, isn’t a matter of sockets. It’s the onboard memory controller.

Can first-generation Hammer memory controllers handle DDR2? Every indication up to now says, “No.” Take away the advantages of the onboard memory controller, and you are effectively left with an extremely expensive Athlon XP.

Objection Four

Prescotts will use 90nm technology. Normally, this is very good news for overclockers, but current indications are that Prescott will have an extraordinarily short useful life for a generation of Intel processors, only about a year, with extraordinary little ramping compared to the typical generation of Intel processor.

We suspect Prescott is the last gasp of relatively cheap current CPU manufacturing, and Intel knows it. They’ll squeeze out what they can, and bite the bullet and start using weird science with Tejas.

We suspect Prescotts aren’t going to a whole lot faster than Northwood for these reasons.

Let us apply the same principles we applied to Intel to AMD.

These CPUs are being sold at the tail end of 130nm technology fortified by one weird science called SOI. To put it mildly, AMD has had considerable problems getting this to work decently. 🙂

They’re just not going to get to speeds much higher than 2.4GHz; if they could, AMD would be planning to sell more models at higher speeds. That’s not bad at all under the circumstances, but why scrape for modest improvements when a newer process with much more potential is not too far away, especially when you’re not terribly hurting now?

Somewhat different reasons, same results.

The Price Crunch and Same Problems, Same Answers…

Same Problems, Same Answers

We’ve talked about Prescott problems. We’ve talked about Hammer problems. For the first time, we’ve comprehensively put them together.

Pretty much the same problems, aren’t they?

So if both companies are having the same sort of problems, and both sets of problems are not very good for overclockers, what’s the obvious conclusion to be drawn?

Neither is much good.

On the whole, if Intel can tame the heat, the situation is somewhat less bad for them, but you can’t say more than that, and that’s a huge “if” at the moment.

And if you can’t see beyond the CPU Cold War “He is who is not with us is against us” mentality, that there is no possibility that neither side is going to be terribly good, and that someone can quite rationally be on neither side, then your brain is broken.

And if you can see beyond, “Well, you have to be in favor of something,” that isn’t evaluating. That’s marketing.

An even closer comparison than Prescott to the first-generation Hammers is Intel’s Willamette. We didn’t like Willamette at all. We even made fun of it, calling it “Willie” all the time.

The reasons why we didn’t like Willamette are almost an exact fit for the reasons we don’t like the first-generation Hammers. The only major difference was that Intel was bold and brazen about Willamette, while AMD looks more dazed and confused.

Throughout the lifetime of Willamette, we had one simple, clear piece of advice: Wait for the next generation.

That wasn’t bad advice, was it?

So when we see the same situation on the AMD side, guess what? We’re going to give you the same advice about it.

Why is that so hard, even impossible for some people to understand?

It has nothing to do with being on one side or another. You don’t have to be on one side or the other.

We’re not here to get you to buy, we’re here to get you to buy wisely. That doesn’t necessarily mean following our advise slavishly; we can’t address every individual situation. No, the idea is for you to think through what we have to say, and decide after having at least considered the points we raise.

Discussion

Leave a Reply

Your email address will not be published.


*