Intel has to make different wafers for products that use different numbers of cores or different GPUs. How inefficient is that?!?! What does Intel do with all the bad processors that don't meet full product specs with low double digit yields?
It's actually far more efficient. Intel can still sell a quad core with one bad core, and fuse it down to a dual core, completely transparent to the end user. But they can also sell a native dual core with much smaller die, and manufacture more of them per wafer, thereby decreasing average costs, and increasing ramp throughput. Furthermore, the choice of two native graphics configs on their dual core means that they can lower their average costs further, every time they only need to ship GT1 graphics.
As for what you do with bad die, most die are recoverable, especially for defects in the cache, which has insane amounts of redundancy. But as I said, even the CPUs are recoverable, and I would be surprised if Intel couldn't also downbin a GT2 with defect in one of the EU processors, and bin it as a GT1.
Therefore, most die should be recoverable, and for the few that aren't, those are scrapped and disposed, as you would expect. But when you see Intel with ~65% margins, you can probably infer by that that they are vastly superior to their competition in terms of yield recovery. It's also due to better ASPs, but that doesn't tell the whole story.
Intel has to make different wafers for products that use different numbers of cores or different GPUs. How inefficient is that?!?!
The cost trade-off of NRE/maskset/overhead vs lower unit cost and high manufacturing capacity depends on volume. Intel's volumes make this a no-brainer for mainstream products.
What does Intel do with all the bad processors that don't meet full product specs with low double digit yields?
Intel would never release a device with anything approaching "low double digit yields" to production even for a 4 figure ASP. Any thing else you need help with?