Semprons are supposed to be introduced August 17.
Unusually, this product lines will incorporate two and eventually three different platforms. First, we’ll have socket 462 and 754 Semprons, later will come socket 939 Semprons.
Based on emails I’ve been getting, this has led to what seems to be a lot of wishful thinking about these chips.
Two Beliefs
There seems to be two beliefs out there about Semprons.
Both points are rather dubious for a number of reasons.
Two Architectures
AMD has two core CPU architectures in place today: the Athlon architecture and the Hammer architecture.
The Athlon architecture consists of a CPU with no memory controller.
The Hammer architecture is functionally pretty much an Athlon processor with an on-core memory controller. (x86-64 is a relatively minor add-on).
This means that any Athlon motherboard must have a memory controller. Not suprisingly, they all do.
It also means that any Hammer motherboard does not need a memory controller; it would be at best superfluous and at worst counterproductive. Not suprisingly, no current Hammer motherboard does.
This leaves AMD three logical choices in any Sempron configuration:
If AMD chose option one, they would have a real problem getting the memory controller on the chip and the one on the motherboard to play together in the socket 462. One or the other would most likely have to be disabled. Disabling the motherboard controller would be the likelier candidate, but that may not even be possible with current Athlon motherboards, and at best, it would likely hurt performance substantially.
I suppose they could mandate that socket 462 Semprons were only compatible with a new generation of Athon motherboards with no memory controller, but that would hardly be a good compatibility move. Besides, a socket 462 CPU with an onboard memory controller would likely perform close enough to a Hammer to encourage many to buy that instead, which is not exactly AMD wants to do with its cheapest chips.
If they chose to disable the onchip memory controller for socket 462, they’ve essentially wasted considerable die space for absolutely no purpose.
If none of the Semprons came with a memory controller, that would be fine for socket 462, but then Semprons wouldn’t work in any motherboard without a memory controller. Mobo makers would certainly balk at adding a memory controller to all their current Hammer motherboards just to handle some cheap chips.
Why Reinvent The Wheel?
The reality is AMD has two perfectly good architectures capable of handling their respective tasks without any engineering reworking at all.
So why reinvent the wheel, especially if the only reason to do so is the “need” in the minds of some for semantic consistency?
Where does it say that a company has to use a single architecture for a product line? After all, we’ve had Intel budget CPUs based on the PII, PIII and PIV cores, but they’re all called Celerons.
Keep in mind that it won’t cost AMD a dime to leave the necessary socket A manufacturing lines alone to continue making socket A Athlon XPs Semprons. Anything else will cost them money to no good purpose.
It doesn’t make technical, marketing, or financial sense for AMD to do anything more than slap a new name onto its cheapest socket A chips and keep cranking them out. It’s simple, easy and fits the bill. If they do anything to these chips, it will be to make them worse, not better, but even there, the simple passage of time will do that.
Thinking that AMD is going to make a better socket A when it’s going to need to get Hammer sales going is wishful thinking, and wanting something to happen is no reason at all to believe that something is going to happen. That only works in Disney films.
I can’t say I’m 100% certain this will be the case; AMD does some strange things, but I’d bet dollars to donuts that this is what this will boil down to.
We’ll see August 17.
Be the first to comment