More Fratricide

Add Your Comments

Xbit Labs has an article out pointing to this ECS CPU support page, which has socket 754 processors we’ve never heard of before.

Essentially, they’re lower-speed Clawhammers with 1Mb cache, given 3100+ and 2400+ ratings.

Junkyard Sale or Just Junk Marketing?

Just to get a minor point out of the way, it’s hard to figure out where the 3100+ fits in speed-wise. If it runs at 2GHz, there’s no difference between it and a Clawhammer 3200+. If it runs at 1.8GHz, giving an extra 300PR points for an extra 512K cache seems a bit much.

However, we’ll find out soon enough, and it’s a trivial point anyway for overclockers. It is the 2400+ that’s the real puzzle. If a processor running at 1.6GHz with 512K cache is called a 2600+, what speed will a chip with 1Mb of cache called a 2400+ be? I don’t think they’d dare call a 1.2GHz chip that, so it’s probably 1.4Ghz.

But why would AMD want to sell a 130nm “big” Hammer (remember, this has 1Mb cache) at what is bound to be a price far below what any 1Mb or even 512K CPU has cost up to now, a price competing against the new Sempron?

Sane and Less Sane Explanations…

A Sane Explanation: When You Have Lemons, Make Lemonade

Why would AMD suddently want to sell CPUs that cost rather more to make due to being die pigs for rather less than previously low-end processors and about the same as the new budget line?

The sane explanation for this is that you don’t make any money at all from a chip you can’t sell.

If AMD’s process technology yields a bunch of chips that can’t reach 1.8 or 2GHz under default conditions, they’re waste product under the current regime.
Selling a 1.4GHz Clawhammer would allow AMD to get $100 or $125 for processors they’d otherwise get nothing for.

While that’s good for AMD, and good for Joe Sixpack buying an HP box, that’s not too good for any overclockers salivating over what looks like a bargain out of the blue.

Should these chips show up, don’t be surprised if they end up being the Hammer equivalent of the TBredAs, not nearly as capable as “normal” Hammers. Could be wrong on this, but the best thing to do is not be the first kid on your block on one and let somebody else prove this contention incorrect.

The Less Sane Explanation

The other possibility is that OEMs want such a chip and that AMD will use their waste chips to initially meet the demand. (While the demand for cheaper chips is unremarkable, it’s doubtful that OEMs would demand cheap Clawhammers.)

This may look the same as scenario number one, but there’s a big difference between the two in what happens afterwards. If OEMs want cheap “real” socket 754 Hammers now, they’re probably going to want them later, too.

What does AMD do when they run out of scrap Hammers when that’s the case? Well, they could say “tough, buy Semprons, you cheap OEM SOBs.” Or they could simply make 90nm socket 754 chips.

Voices of Dissent for Socket 754

An increasing number of people are voicing the opinion “Socket 754, now more than ever.” Their reasoning goes as follows, “You’ve all waited for socket 939. It does practically nothing for performance. You’ve seen 90nm chips. They do practically nothing for performance either. Why wait and wait and wait when you can get socket 754 now?”

So far as it goes, it’s not an unreasonable argument. Dual-channel doesn’t do a whole lot with Hammers, and 90nm chips aren’t blowing their older brothers away yet.

However, this argument leaves out the biggest reason not to go socket 754, the primary reason why we’ve been against socket 754 systems all along. Socket 754 was supposed to be shoved into the bargain basement, which made it a dead-end system.

Will AMD stick to this policy, or will they reverse course and rip up their strategy again? Given that they now plan to make 90nm Semprons, will “real” 90nm socket 754s be far behind? Whether OEMs are behind Clawhammer 2400+s or not, they’d probably like that idea.

Should that happen, the last big reason not to go socket 754 vanishes for most typical AMD fans. What’s important here is not the actual emergence of 90nm socket 754 “real” Hammers, but rather AMD announcing such a thing. Once that happens, people can buy a cheap 130nm now, and top it up with a cheap, improved 90nm later on. They’ll lose a bit in performance, but they’ll probably save enough cash to reasonably justify it.

Unfortunately, this isn’t a given. AMD might respond to OEM demands for cheap “real” Hammers by simply enabling x86-64 on Semprons, and let socket 754 Newcastles or Clawhammers expire.

In short, if you buy socket 754 now, you’re gambling that AMD changes its mind. If you win the bet, you get Hammer performance longer for cheaper. If you lose, though, you have a dead-end platform.

Why is this so bad? Well, it’s certainly not bad for those of you reading this with more patience than money and/or enthusiasm for Hammer. Things are going wonderfully for you.

It is bad for AMD because moves like bringing to Semprons to 90nm and selling really low-end Hammers are indicators that AMD’s high pricing strategy is starting to crack, and probably will break open in the next six months or so.

A Strategy Unravelling…

A Strategy Unraveling

Once upon a time, AMD’s strategy for the next year looked simple and clear: Be just like Intel and have two product lines with significant performance differences.

This is going to hell, primarily because AMD can’t distinguish their products enough to justify much higher prices for their higher-end mainstream products, at least not to the people willing to buy them.

They can’t use 90nm vs 130nm technology because 1) there’s not much difference between the two right now and 2) even if a significant difference emerges down the road, they don’t have the capacity to make a lot of 130nm budget chips.

The Hammer architecture itself is too good for AMD’s own good. Its memory controller minimizes the differences between single and dual-channel, and its general architecture reduces the benefits of additional cache compared to Intel designs.

In short, if the budget product does 90-95% of the job, why pay an extra $100-$200 for 100% if there’s nothing else to distinguish between the two? This is an argument AMD fans are very receptive to hearing, and if budget processors end up using the same technology as the expensive stuff, that’s what they’re going to buy.

The only significant feature AMD has left to use to distinguish between product lines is x86-64, and Intel is bound to blow that up by simply enabling it on both PIVs and Celerons. Doing that doesn’t hurt them one bit more.

While all this is happening, socket A has finally reached the end of its useful life, and AMD can’t count on it paying most of the bills much longer.

So AMD is scrambling to do what it can do with what it can and cannot do, and meanwhile, you cheap SOBs out there (I’m smiling when I say that), both those who buy and make the computers, just keep waiting and maybe grinning a lot more than you did a while back.

In short, over the next six months or so, AMD is going to have to start selling a lot of Hammers. The customers inclined to buy them don’t want to pay much for them, and they’re more than willing to forfeit 5-10% performance to save $100.

It’s turning into a buyer’s market, which means that AMD will probably be right back into the pricing quicksand.

Fratricide

With one little exception, who have I not mentioned in this article? Intel. This is pretty bizarre given that we’re talking about CPU sales.

What is so queer about all this is that AMD isn’t primarily competing against Intel. It’s competing against itself. People aren’t thinking “Should I buy Intel or AMD?” They’re thinking, “What’s the cheapest AMD system that will give me the most bang for the buck?” Socket 939’s toughest competition doesn’t come from Intel; it comes from socket 754, and if AMD levels the playing field between the two on upgradability, the cheaper option is going to win.

That’s what “fratricide” means: Killing someone close to you.

The folks at Intel must be laughing their heads off. They’re on the ropes, and AMD responds by pummeling themselves in the face. As Napoloen once said, “. . . never interfere with the enemy while he is in the process of destroying himself.”

Yes, different products for different market segments is fine, but you shouldn’t have multiple products fighting for the same people (and confusing the hell out of them in the process), especially when the one that gets you less money is the one the audience is likely to pick.

Until a month or so ago, the AMD approach was to make people spend more for their next processor. Their recent moves will end up doing the opposite, pushing people down the line.

Good for you, but not good for what is still a pretty fiscally fragile company.

Email Ed

Leave a Reply

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