This will not be a fun day for AMD fans.
There are plenty of reviews out there, and they all say the same thing: the Thoroughbreds don’t overclock very much.
Well, I hate to say we told you so, but we told you so, over three months ago.
There’s an obvious, and not-so-obvious problem.
The Obvious: Heat
The obvious problem is that this is one hellaciously hot chip, which must be a first after a die shrink.
Now how can that be?
When you pump 15% less electricity into a chip that’s 40% smaller, the chip gets hotter.
Let’s take the typical thermal power of a 2100+ Palomino vs that of a 2100+ Thoroughbred.
64.3 watts/128 sq. mm = .502 watts per square millimeter
56.4 watts/80 sq. mm = .705 watts per square millimeter
40% more heat to be dissipated per square millimeter.
Here’s the real cringer:
64.7 watts/120 sq. mm = .539 watts per square millimeter
This brings us at least right back to the Thunderbird days in typical overclocking situations, if not worse.
Not-So-Obvious: The Hammer (Black) Hole
The initial results are not in themselves a matter of great concern. You may recall that the Thunderbirds improved tremendously with continued refinement and tweaking.
No, the matter of great concern is that AMD is going to treat the Athlon as a cash cow, and not refine and tweak it much or at all, at least for a while, or maybe not until UMC takes over.
Here’s an indicator of just that.
“Richard Heye, AMD Vice President of the Platform Engineering and Infrastructure Computation Products Group, explained to us in a no-nonsense fashion that we would not see a 166MHz bus on the current CPUs any time soon, as resources for implementation had been pulled to be utilized on Hammer R&D.”
Folks, this is not hard for a CPU manufacturer to do compared, to say, lowering the power requirements of the CPU. Provided you have the right equipment, running at 166MHz FSB is no big deal with current XPs.
There are plenty of indications Hammer has become AMD’s black hole, sucking up all its material and human resources.
Computex sure showed us Hammer sucked up all of AMD’s PR resources. 🙂
More seriously, promised improvements have been slowly but surely stripped away from AMD’s plans. SOI got removed from the first Bartons.
The Bartons themselves are being outsourced, for God’s sakes. If you’re an ambitious AMD engineer, do you really want to work on improving something somebody else is going to end up building, or do you want to work on the project that’s the company’s future?
If it hasn’t happened already, one day soon, a few Intel engineers are going to rip one of these Thoroughbreds apart. And I bet their reaction is going to be that this die shrink was a rush job, with the absolute minimum done to get it out the door. That wattage requirement is a red flag.
Something is happening over in Dresden.
You have three possibilities:
UMC’s problem in six months or so.
I think it’s the third.
What’s An AMD Fan To Do?
If you have a current system, wait either for Hammer or for whatever UMC ends up doing with Athlon, or go Intel. What else are you going to do?
The strategy for any new socket A system is simple. Lower the multiplier, increase the FSB. Shoot for 166MHz or better.
Forget asynchronous memory settings on socket A platforms; the memory bandwidth ends up exceeding the FSB bandwidth.
One of the few fortunate features of the Thoroughbred is that along with Thunderbird heat comes Thunderbird ease-of-unlocking. It looks like the pencil will rise again!
Keep in mind all these chips will eventually become the new Durons, so it is probably wiser if you have to put something together now to go low-end, perhaps sacrifice a hundred or two MHz, and await better days in 2003 with Barton. If you have a current system, of course, you can always wait until they become Durons.
For now, though, and if the Hammer black hole continues to suck, perhaps for the next six months or so, the overclocking action looks like it’s pretty much going to shift to Intel.