| Bofferbrauer2 said: Who said anything about scaling? It seems you didn't get the point. It's just way too early yet to move to 7nm. Your link is from mid 2016, when the 14/16nm processes where already very well established. But for the 7nm process, we are at a point like mid 2014 was for 14/16nm processes: Too expensive yet for mass production, but tech is getting tested with some early designs of future chips. For reference, your article is from June 2016, almost a full year after Apple came with the A9 in the iPhone 6S/S Plus or Qualcomm with the Snapdragon 820, and over a full year after the Exynos 7420, all having been produced in 16/14nm processes, meaning the process was already ripe at their releases. What he's saying in the article is anybody who hasn't made the jump yet will be left behind because there's no reason anymore to produce high performance chips in 20/22 or even 28nm to avoid bad yield rates (and thus high prices) anymore. Oh, and Sony isn't designing anything, they are just asking the chip manufacturers if they can produce a chip which is able to do what they ask (and pay) for. So if it can be done cheaper with GDDR5, they will take it no questions asked. |
Your point doesn't make much sense since foundries determine what is and isn't viable for mass production ...
And it's NOT too early to move to 7nm since console APUs will be the size of a smartphone SoCs at that point so cost concerns are mitigated with smaller dies. You also misunderstood the article too, it only argued that you should transition to FD SOI from FinFET to reap more benefits. All home console manufacturers still saw reasons to use 14/16nm even though it's permanently more expensive per gate either way ...
Sony still has to redesign the APU to use 7nm cell libraries ... (you can't reuse the physical design from older process applied to a new process since they don't have the same design constraints)
