Depending on how well the new DX12 Multi GPU modes go over in terms of support, multiple small die could once again be a viable strategy.
I don't understand how several posters are trying to push this theory when:
1) AMD has moved away from small die chips looking at the general trend starting with HD3870. What is the trend in die sizes?
UP, it's not down. So why in the world would AMD suddenly say Jeez, let's go back to small die strategy? It makes no sense because (1) It contradicts what they have actually been doing in reality and (2) it contradicts the entire foundation of Lisa's strategy to move AMD away from a budget brand. Budget brand <=> no single large die chip flagship cards like 3870->6970 days.
2) With delays of Fury X2 and multi-GPU issues in some popular AAA 2015 games (even though a lot of these are console ports), it's too risky to rely on dual-chip mid-range cards as your strategy. This is even more dangerous because UE4 doesn't support multi-GPU out of the box. This doesn't mean that AMD won't make multi-GPU chips for VR, but relying on small die multi-GPU chips for gaming will fail automatically. This is because if the competitor's product offers 40-50% more performance against a small die chip, no one is going to buy a card that requires CF to get that extra 40-50%. This has already been proven over and over with HD4870X2, HD5970, 6990, 7990, R9 295X2, etc. Dual chip GPUs hardly sell and most of the market doesn't care about them for gaming. Why would AMD make a crucial mistake of going against market trends? Considering they delayed Fury X2, they are actually paying attention how these style products are mostly not very popular.
3) It's actually not even possible to achieve an 80-100% gain in performance over Fury X when coming off a 596mm2 chip with a 300-350mm2 14nm AI. This alone destroys the theory of the small die strategy. What would be the point of just improving perf/watt 2X of Fury X but be barely faster? Perf/watt must go hand-in-hand with increased performance or otherwise such a strategy is again an automatic fail because the competitor will be doing BOTH.
4) Small die strategy implies NO flagship performance in a single chip card. That means, you cannot charge flagship GPU prices. That means you have lower profits margins, lower revenue and automatically give the competitor the entire $500+ GPU market. Every one of these things is what AMD is trying to fix while small die strategy goes the complete opposite way of higher margins, higher revenue, better brand image, etc.
5) With HD7970->290X, we have seen AMD roughly use the last gen's flagship performance as next gen's mid-range. This is not a viable strategy when utilizing the small die strategy since a mid-range small die chip of one generation cannot again be reused as a mid-range small die chip of the next gen -> You also have no room for the upgrade on the next node.
Since 14nm node is likely to stick around for years to come, it would be impossible to make substantial gains of 25-30% in performance if you are stuck on 300-350mm2 14nm die.
We just have to look at 7970->290X->Fury X to see that as long as there is no next gen node, large die strategy is not just an option, but it's going to become essentially to remain competitive since new nodes are not gong to be available to AMD like clock work every 18-24 months as was the case in the past.
My opinion is that small die vs. large die strategy isn't even debatable. The
only way for AMD to turn around its GPU business is to fully adopt the large die strategy and scale its chips from 300mm2 to 600mm2 on the same node, and then start the entire cycle over with the 10nm node, then 7nm node, etc. We have already seen both companies doing exactly that with 28nm node, and there is 0 reason to indicate this is suddenly going to change, especially when Fury X has come so close to capturing the stock performance crown this gen.