When You Have Lemons, Make Lemonade . . .

The Inquirer reports that AMD is going to have to downgrade the already low speeds on its initial Phenoms. No more than 2.4GHz for you, and even just 2.4 will have a 125W rating attached to it.

This is getting ridiculous. “True quadcore” is truly becoming a quagmire.

It’s becoming obvious that AMD is not going to be able to field a competitive desktop quad anytime soon. They need 3.2-3.4GHz to be more or less even with a Penryn. Even if you believe the current AMD roadmaps, they’re not even promising that.

What adds insult to injury is that while AMD can’t make competitive quads, they won’t even try to make competitive duallies.

Maybe its time for AMD to swallow a chunk of its dwindling pride.

On the desktop, quadcore will remain more a prestige item than anything functionally useful for quite some time to come. Look at the comprehensive benchmarks comparing duals to quads. Few of them get any real benefit from the two extra cores.

Quadcores also warp the power parameters. Whether they’re actually used or not in real life, systems must be designed assuming that they’ll be used. Even for Penryn with its high-K and halfnium, speeds are being stunted by power constraints. While AMD probably has more problems than that, given the 125W for a 2.4GHz processor, power constraints are if anything even a bigger problem.

Maybe desktop quadcores are a luxury AMD can’t afford this go-round. Maybe AMD should just give it up and try again at 45nm.

I do not suggest AMD should give up quads altogether. They can keep making Barcelonas for the server market. High speeds aren’t as important, they’ll get paid better for them, and quads actually offer real benefits over duallies in the mainstream server/workstation market.

But for the desktop? Forget about it. Aim for an easier target. Whatever yield problems AMD might be having, they’ll certainly be improved if the object is to make two cores rather than four.

Speeds can probably be made a bit higher if the power envelope covers two rather than four processors. That will make K10 processors more competitive today and tomorrow.

It would be better and more truthful if AMD’s marketeers could show K10 duallies matching or beating Penryn quads in real-time apps and games than having quads beating slower Intel processors.

Yes, AMD will get less money from a dually than a quad, but keep in mind that AMD will get to sell at least two duallies for the same die space as a quad (and if quad yields are really that bad, maybe more than that). Intel’s expected pricing has two fast duallies costing more than one slower (12MB) quad, AMD could expect the same.

If you’re shaking your head at this, consider this. A couple weeks from now, AMD is going to release its slow Phenom X4s. A 2.4GHz Phenom X4 is going to get beaten like a rented mule. The Penryn 3GHz will beat it all the time, the fastest Kentsfield will beat it almost all the time, the fastest Intel duallies will beat it on the large majority of non-quad sensitive benchmarks, even the fastest X2 duallies will beat it for the same.

The only way AMD can even attempt to duck this disaster is to prohibit these comparisons, which is a cure worse than the disease. That will just drive the comparisons into the “underground” forums, and Phenoms will end up earning a Vista-like reputation.

And it won’t get better, if AMD is really lucky, until next June or so? How is that better for AMD?


Be the first to comment

Leave a Reply