14 DAY TRIAL // The early bird gets the worm. This adage is one that tends to hold true in all cases, as the “first come first served” way of attending the needs of customers is usually the one that causes the fewest headaches and paperwork mishaps. Unfortunately, NVIDIA and AMD are, for once, on the wrong side of the field.
Maybe we shouldn't be all that surprised to learn that Apple and Qualcomm are disrupting the marketing plans of chip makers only peripherally related to them, in the case of graphics anyway.
Still, it's an odd irony when NVIDIA and Advanced Micro Devices both have to wait in line because the other two managed to secure the first batches of new-generation chip wafers.
More amusing is that Globalfoundries practically belongs to AMD, even if that's not true on paper, but AMD has to wait for the others to lay off on the 20nm orders a bit.
No AMD 20nm chips pushed back to April or May 2015
The delay to AMD's roadmap is the smallest of the two, of just two months or so. This is because chip shortage is solely owed to Apple and Qualcomm reserving most initial 20nm batches, as we said.
Originally, though it was still the rumor mill that released this info, AMD was supposed to launch its new Hawaii GPU, or whatever its name would have been, in March next year. Moreover, the Hawaii was going to be followed by an unlocked Tonga and a Fiji XT.
These things will still happen, but later, between April and May, and an interesting note is the name of the architecture: the 20nm AMD GPUs will be based on the Caribbean Islands.
The WCCFtech report could be false we suppose, but the source is said to be the same one that revealed AMD's switch from 28nm to 20nm, NVIDIA's decision to make GM204 GTX 980/970 video products on 28nm instead of 20nm, and even the skip from 28nm to 16nm.
Which brings us to the much greater delay that NVIDIA's graphics processing unit has been subjected to.
NVIDIA won't move to next generation GPUs before 2016
Presumably, NVIDIA won't bother trying to once again redesign the Maxwell architecture for a new node. Originally, Maxwell was made for 20nm, but after TSMC failed to implement it and then skipped it altogether, NVIDIA remodeled it for 28nm.
Making 16nm Maxwell shrink might have been doable for 2015, but that's supposedly no longer part of the plans.
While NVIDIA will still launch the GM200 Maxwell GPU in 2015, the chip will still be a 28nm unit.
Come 2016, the Santa Clara, California-based green company will finally have all the 16nm chips it needs from TSMC, but none of them will be Maxwell GPUs. NVIDIA will jump straight to Pascal instead.
