Now that AMD has announced that they will indeed sell three-core processors, let’s go into a bit more depth as to why this isn’t the greatest thing since sliced bread.
The problem isn’t that three cores are inherently bad, stupid, foolish, etc.. The problems with tri-core are the circumstances under which they’re being launched and the competition they’re likely to face.
Supply and Demand
There are three ways AMD can end up making a tri-core:
This isn’t an either/or; if I had to bet, I’d bet that AMD will do all three.
Let’s look at each option:
Using faulty chips: Selling chips you’d otherwise have to throw away (or maybe make duallies from) might seem to be an unadulterated plus, but if the public perception becomes
“Tricore = Selling a broken chip” (an idea Mr. Intel might like to promote), this could turn into a public relations disaster that could hurt ALL AMD sales.
Imagine going out to buy a 12-pack of beer and seeing one brand selling for 25% less, with three of the cans marked, “Do not drink: Stale.” Don’t you think at least some people will get really turned off by that, even if the stale beer does them no harm? It’s one thing to cripple part of a healthy chip to make it a bargain; it’s another to sell a chip crippled because it couldn’t cut it whole. People can be funny about such matters, and even when they’re still willing to buy, they’re likely to want a very cheap price for it.
Unless matters have gone completely to hell at Dresden, it would be pretty perverse to have a product line depend on a poor manufacturing process. What if Dell and HP put out tri-core product lines? What are you going to tell them if they want more, “Sorry, we didn’t screw up enough, we promise to do worse next week?”
Of course, AMD isn’t that crazy, so they’ll also do one or both of the other options.
Disable a healthy core Incur all the expenses of making a four-core, then deliberately sabotage it so you can sell it for less. This is just what a company staring bankruptcy in the face needs to do.
Yes, GPU companies regularly do something similar, and even Intel (very quietly) has been known to make a few chips that way, but Intel and nVidia aren’t facing bankruptcy. Moreover (for different reasons), neither company is capacity-constrained the way AMD is. 65nm K10s are big and chew up a lot of die space. When you have just one functional 65nm fab, and your second source hasn’t done much for you yet, this looks to be a waste of resources at best.
Again, the big OEMs could end up putting AMD deeper in the hole. Let’s say they like tricore and order lots and lots of them. AMD doesn’t dare tell them “No,” so they may find themselves breaking millions of quadcores for a pittance while not being able to supply the little people with unbroken ones.
I don’t think AMD’s crazy enough to do that as a matter of regular policy, though if they get OEM demand wrong, they may have no choice but to do so. That leaves option three.
Come Up With a Tricore Design This is the least bad option insofar as production is concerned, which is why this is likely to eventually become the main source of such chips. Keep in mind, though, that other companies cripple their chips precisely to duck multiple designs (and their costs). In any event, this just brings us to the other side of the equation.
What Is Intel Going To Do?
AMD has the bad habit of making plans as if Intel didn’t exist. It’s not just a matter of anticipating what Intel might do, they don’t even respond to what Intel says they’re going to do.
A good example of that was dual-channel memory. Intel publicly announced that six months before Hammer was going to come out, and AMD said nothing until the initial release, when it was “Whoops, we’ll do it, too, wait another nine months.”
I don’t know what Intel might specifically come with to counter tricore, but I’d say it would be more likely than not that they’re going to stick with their announced game plan, which is to make 45nm quadcores mainstream in 2008.
AMD’s claims that the average computer user isn’t “ready” for quadcore are just silly; price it right, and they will come. They tossed out a factoid saying that only 2% of the market was quadcore, but that measurement went back to product introduction in November, 2006. From November to July, those quads cost over $500, and I think it can be safely said that marketshare wasn’t even close to 2% during that time.
But when the price went to $266 in late July. 🙂
$266 is still pretty high for a mainstream purchase, but that’s what 45nm is for. By the middle of next year, and probably sooner, we’ll see sub-$200 quads from Intel, let’s say $180.
And when that happens, and presuming Penryns will continue the expected lead over K10s on the desktop, what’s a tricore going to end up costing? I would venture a price range of $90-150 (and less, of course, for the OEMs). Might be good for you, but not good for the company.
If AMD were going to get a lot of money for these things, we’d have a much different ballgame, but that’s pretty unlikely.
AMD needs to start making a lot more money per CPU, no ifs ands or buts. If they don’t start doing that pretty soon, they aren’t going to be around much longer. Unless quadcore yields are horrendous and this is a forced move (which is pretty bad news by itself), tricore is likely to end up being another shot in the foot for the company and its finances.