AMD Phenom II X6 1055T 2.8GHz 95W

[ridney]

All that matters is die size. That translates into $. You save some mm^2 on the core and lose on cache or the other way around, it doesn't matter if the end results are similar and one performs much better than the other.

here's what JF had to say about this matter, he explains it very clear and quite convincingly

Well, if you are going to go down that path, where do you stop? What about profit per wafer? You get significantly more profit out of a MC wafer than you do an atom wafer for instance. Or any of the other intel client wafers. So by that logic, is intel making a mistake by making those parts? Absolutely not.

In the fab business you maximize your revenue based on getting more wafer starts, driving down your cost per wafer and by driving more revenue per wafer.

In reality, the number of server wafers vs. client wafers probably makes this a moot point for both intel and AMD.

Take Intanium for instance. Higher wafer cost (lower yield because that product is behind the process curve), higher development costs because it is a more expensive product to build and support with much lower volumes.

I can guarantee you that if you look at the true (fully burdened) cost of building an Itanium that sells for $1500 and an Opteron that sells for $500, that we are most likely making far more profit on the Opteron. They might make more revenue, but their costs are going to be a lot higher.

Then take the xeon line in general. 3000, 5000, 7000. Three different products to develop. Three sets of R&D. Three dies. Three sets of masks. Thress sets of different testers (one for each socket.) Three pieces of silicon to manage both in process and finished goods.

Then look at Opteron. 4000, 6000. A single die to manage. A single die that I can hold in the manufacturing process until the end to determine which product it goes in. You do realize that from die to finished good is only ~2 weeks. So by utilizing the same die, I can make a lot of final inventory decisions ~2 weeks before the product needs to be in a customers' hands. With 3 different discrete parts, all Xeon choices need to be made ~13 weeks ahead at the start of the process.

One die to develop. One die to manage. One die to process. And only 2 sets of testers. Inventory to minimum, cost to a minimum.

If you look at per die cost, you could argue that intel has an advantage because of the die size.

But if you want to look at fully burdened cost, our costs are a lot lower to get the Opeteron product to the streets than it is for Intel to get the Xeon product to the street. Every time someone argues that intel has "10X the resources of AMD" they are helping make the argument, because resources don't work for free - I know I don't and I am pretty sure everyone reading this wants to be paid for their work (and probably wants to be paid more.) Die cost is only one dimension of the total cost equation and as long as you don't look at the full picture, you will never see the actual impact.

We were able to compress the pricing and remove the 4P tax because of those efficiencies I just talked about. And while a lot of the intel fans seem to think that paying more for a product is a priviledge, when you sit down across from a real live paying customer, they disagree. Believe it or not, they actually appreciate paying less.