Nvidia CEO Jen-Hsun Huang Jensen recently told investors that the transition to 28nm is being executed better than 40nm, but despite this rather positive claim it doesn't mean that 28nm is in perfect, or for that matter great shape. Our multiple sources that are close to manufacturing circles are claiming that 28nm is more complicated than most punters expected.

Basically 28nm graphics won’t be that easy and 28nm mobile chips in 2012 won’t have great yields. The problem is not limited to one fab, both TSMC and Globalfoundries have the same problems with good yields on 28nm, meaning both AMD and Nvidia will probably be affected.

This means that graphics chips will ramp much slower than many expected and that despite the AMD launch of Radeon HD 7000 series in this year, they will most likely suffer from shortages for quite some time. This is probably one of the reasons why Nvidia wants to wait a bit with its 28nm chips, since it is likely to face the same issues. With all this in mind, we really wouldn’t rush to call the transition easy.

Just remember that the whole mobile industry relies on 28nm Cortex A9 cores in 2012, and that the yields and manufacturing cost will play a big role in mass availability of these chips.

Unfortunately for the rest of industry, Intel makes much smoother transitions, but we also got an explanation for that. Chaps close to GloFo told us that Intel only have to optimise process for Intel's chips while TSMC has to make it work for dozens, if not hundreds of customers. GloFo also has more than just AMD as a customer and has to optimise its 28nm for many other partners, which is no easy task.

In any case, 28nm is set to kick in next quarter, with AMD's Radeon HD 7000. Nvidia will follow in early Q1 2012, while the mobile phone and tablet chips are expected a bit more later in 2012. AMD will introduce its first 28nm Krishna APUs sometime next year, but we still don’t have an exact date. It is worth noting that AMD pioneered 40nm GPUs with the HD 4770 back in April 2009.  It is safe to assume that 40nm GPUs in will continue to roll of assembly lines three years on, thanks to a number of issues, but that is another story altogether.