There’s an AMD Tech Tour going on. Somebody from a website went to it. Part of the tour was a technical conference.
The item which should stir your interest was what was said about Hammer’s memory controller.
To get everybody up to speed, most of the performance improvement Hammer will have over current Athlon technology will be due to a memory controller integrated into the CPU. This will improve performance by reducing latency.
Memory controllers have previously been a mobo issue, controlled by the mobo maker. Outside of the AMD chipsets, third parties designed the chipsets and the memory controllers and did with them what they liked.
Now this will be a CPU issue, which AMD controls.
This raises the question: Just how flexible is the memory controller on Hammer for future memory improvements?
This really breaks down to three questions:
1) Will the initial Hammer controllers be able to handle DDR400?
The answer to that seems to be unclear (though this question is relatively unimportant compared to the next one).
However, it’s important to note that making the memory controller means controlling (most) overclocking.
If AMD wants to throttle Hammer overclocking, they can do so by limiting the memory controller. Not saying they will, I’m saying that this gives them a tool with which to do so if they choose, and one that’s buried deep inside the CPU.
Remember that overclocking doesn’t cost a chipset or mobo manufacturer any money. If they can sell more mobos with overclocking-friendly settings, that’s an incentive to provide them.
AMD is bound to have a different perspective on the matter; at the least, it’s certainly not a purely “win” situation for them like it would be for, say, Abit.
If the initial Hammer memory controller isn’t designed to handle, say, 200MHz memory speeds, it’s at least questionable whether a workaround will be easy or even possible.
Again, there may be nothing to this, but it’s a good question to keep in mind as more Hammer details become available. The place to look will be the FSB and memory specs on Hammer boards. If they pretty much let you do what you want, no problem. If they don’t, there’s something to this.
2) Will the initial Hammer controllers be able to handle DDR-II?
The answer to this one appears to be “No” and if true, this looks to be the real bad news. The initial Hammers are reported to have a DDR333 interface. This is not comforting when you see stories about DDR-II modules coming along very quickly. If you believe what Samsung
says in that story, they’ll have DDR-II 533MHz modules available around the time the first Hammers comes out with a DDR-I 333MHz memory interface. Even if that Samsung estimate is overly optimistic (Update 5/31: Overly optimistic by a year, it turns out, it’s supposed to be 3Q 2003), there’s hardly an incentive to buy an initial Hammer. Why buy a CPU that’s going to be obsoleted very shortly by something providing much more bandwidth, which the Clawhammer could certainly use? Update: This is no longer applicable given the change of date for DDR-II. However, by the time Clawhammer comes out, dual DDR PIV boards will be commonplace.)
How long will it take AMD to come up with a Hammer that can handle DDR-II?
3) Once Hammers get to a DDR-II standard memory controller, how flexible will that be?
Once we’re in a DDR-II environment, again, how flexible will that controller be? When we go from 533 to 666, or any step beyond that, will that mean a new CPU, too?
It’s true that as a practical matter, most computer owners would practically be unaffected by this. Outside of overclockers and hobbyists, people aren’t swapping out mobos and memory to new standards with any regularity.
However, starting what is supposed to be a high-end performance processor line with what will quickly become a low-end memory controller certainly seems to be starting off on the wrong foot. Especially if the Hammer-Athlons are using only single-channel DDR.
It looks especially bad when by that time dual DDR will certainly be an option on the Intel side.
Speaking strictly for myself, a personal computer owner, I don’t see why I should buy an initial Hammer system just to repeat the whole process again six months down the road, or whenever AMD feels like handing me what I thought I should have gotten in the first place. I know 2.7Gb/sec is a bit of a bottleneck even now; 4.2Gb/sec sounds better, or at least longer-lived. I’ll just wait.
And I suspect I won’t be alone if the answers are what they appear to be.