A9R4 is a lot more than a simple bug revision - unlike most revisions it improves performance significantly due to a new branch predictor and a new prefetcher. The IPC gain coupled with the frequency boost means the gap with A12/A15 is now fairly small.
Samsung is also in the position of already having gained market share for its Galaxy smartphones using a licensed ARM core approach while being able to compare its position against Qualcomm Snapdragon (which it uses) and Apple A series processors (which it doesnt).
Well said, MicroMan. I think in any industry, the creative types -- quite appropriately -- aren't really focused on costs of their endeavors, just on end results. Upper management earns the big money for making those calls on the risk/return ratios, which sometimes seem incredibly insightful or hopelessly stupid to the rest of us after the results start flowing in. We often see a CEO fired after a product fizzles; it's their job to get it right. So here's a rare moment of praise for the CFO and CEO who have the guts to say: "Yeah, go ahead, spend the money. What you guys are working on could be a winner for us."
There are so many things that make a successful processor chip. Power consumption is critical, but selling the chip into a outrageously successful end product is the most important. Maybe 10-20% more battery life makes a modest product an outrageous product, but probably not. But the cost of manning your own design team to squeeze out less current, then possibly ongoing software differences, could also kill the financials if the end product doesn't hit volume. Since Samsung sells the end product (smart phone) it *should* know it's customer the best and have a better confidence in whether those volumes will come. Not for the faint of heart.
I think the big thing that got a lot of the mobile SoC guys was that the A15, while fast, used too much power, and the A9 was getting outdated. It looks like the A12 will fill this gap, but it's not going to be available for awhile.
I understand what Nvidia is saying, and I would agree that with ARM putting out so many cores the advantage of doing a custom core design would seem limited. But I also agree with Peter. 15 percent is something. That probably makes it worthwhile for Qualcomm.