According to a report from WCCFTech, both AMD and NVIDIA have delayed their next-gen GPUs – entirely due to supply constraints on 20nm and 16nm respectively.
The GPU industry has been stuck on the 28nm process for an unprecedented 3 years now. While 2014 did bring TSMC’s 20nm process, the last 3 years has seen a massive proliferation of mobile devices, and much of the early 20nm capacity seems to have headed the way of the likes of Qualcomm and a major new client – Apple. Flagship GPU dies are also much larger than mobile SoCs and take longer to hit acceptable yields. As a result, AMD and NVIDIA are left with a limited supply of 20nm dies.
Anticipating this shortage, NVIDIA chose to release GM204 on 28nm, instead of waiting for 20nm. WCCFTech’s sources claim that NVIDIA has chosen to skip 20nm altogether and move straight the 16nm. Unfortunately, it looks like 16nm will have its share of supply issues as well, delaying NVIDIA’s next-gen GPU all the way to 2016. With such a delay, we could see GM200 release at 28nm, with next-gen Pascal straight at 16nm.
Meanwhile, AMD chose to release the relatively minor refresh Tonga at 28nm, with the true next-gen flagship Caribbean Islands GPU still scheduled for 20nm. Previously expected to release in Q1 2015, AMD’s next flagship now appears to be delayed to Q2 2015, once 20nm supply stabilizes. AMD has publicly announced that they intend to stick with 28nm for a majority of their line-up across APUs and GPUS, so 20nm would likely be restricted to the high-end GPUs. We could see the Caribbean Islands family feature a mix of 20nm and 28nm GPUs.
It is still early and there seems to be a lot of possibilities, particularly for NVIDIA. However, it seems clear that both AMD and NVIDIA have been developing their GPUs around TSMC’s plans in the last couple of years, and the trend seems set to continue.