Originally Posted by ILoveHighDPI
It's like an Olympian who is capable of obliterating a current world record, but since their sponsor gives them a million dollars every time they break the record by any amount, it's in their best interest to only break the world record in the smallest increments possible.
Intel could probably be getting 20%
performance gains on a yearly basis, but that would be what's best for consumers, not what's best for Intel.
I guess I don't inherently disagree with the best CPU manufacturer in the world making lots of money for being the best CPU manufacturer in the world, but as an enthusiast it's kind of painful to know that everything on the market right now could be way better than it currently is.
Don't underestimate the need to verify designs. The hardest parts in the design aren't the ideas; it's testing. Once the logic is decided on, as many test cases as possible must be checked. They can sometimes be mathematically verified, but often the complexity dictates that the only viable solution is to test, test, test and then hope that everything was caught (or things like the Pentium FDIV bug or the Phenom TLB bug happen). Once the chip has tapped out, it's too late. That said, the more things that change in an iteration, the greater the chance of creating new bugs. Launched chips all have hordes of bugs, errata as it's called (Here's the i7 900's list
(PDF) which is 55 pages of bugs with lots of them not having workarounds
Once all the known bugs are fixed, the design is tweaked again to better fit the fabrication process, but once again, there are new bugs that crop up causing a new round of testing.
And the testing continues...