During the Q2 financial conference call, AMD's SVP and COO, Lisa Su, assured that their company will out 20nm products next year followed by 16nm FinFET later on. While there have been rumors that AMD might introduce 20nm products in late 2014, I believe that this statement is an end to all those reports since AMD's entire lineup is to stick with the current 28nm process node this year.
AMD Assures 20nm Products in 2015 - 16nm FinFET Later On
In Q1 2014, AMD forwarded the same statement that their graphics and semi-custom business products will both stick with 28nm. We have seen several 28nm products across AMD graphics and processors designs which include the Radeon R9 and Radeon R7 series for graphics while Steamroller, Jaguar powering AMD’s processors and accelerated processing units were also based on 28nm.
"We (AMD) will be shipping products in 20 nanometre next year and as we move forward obviously a FinFET is also important,” Lisa Su (Senior Vice President and Chief Operating Officer at AMD) via Fudzilla
This news is interesting for the moment since AMD is on the verge to introduce new graphics processing units for the consumer industry. These include the Tonga, Iceland and Maui which are reportedly launching in Q3 and Q4 2014. So we can more or less confirm that all three of these chips will feature the existing 28nm process design while their replacements which are now believed to arrive 2015 will be based on 20nm. The same goes for AMD's upcoming semi-custom APUs which are part of AMD's Skybridge initiative which allows pin-to-pin compatibility for x86 and ARM based processors. The Cambridge APU will feature this technology leveraging the efficiency while the K12 ARM core which is a 64-bit ARM custom designed core will use 16nm FinFET technology. It should be noted that AMD also has a true x86 architecture in development to feed high-performance needs which should not be expected prior to 2016-2017 so we may see it based on a smaller node.
The industry would have been on-board the 20nm bandwagon if it wasn't for TSMC to delay the the process by such as large margin. The delay has forced Apple and Qualcomm to look elsewhere leaving TSMC loosing orders from their biggest clients. AMD and NVIDIA will be staying close to TSMC since they rely on 20nm chips for their next generation GPUs but even NVIDIA is being forced to launch their upcoming GeForce 800 series
on the 28nm process instead of 20nm as reports reveal. Coming in Q4 2014, a few months after AMD introduce the latest 28nm replacement to Tahiti codenamed Tonga, the Maxwell generation of graphics cards were supposed to be based on 20nm but this delay tells otherwise. It is all but known that the initial production was prioritized for Apple but if we go back to Kepler's launch, NVIDIA faced a similar situation when the 28nm node was given extra priority for AMD GPUs when they were launched in January 2012 while the Kepler GK104 chip faced 28nm production woes. This forced NVIDIA to launch their GK104 based GeForce GTX 680 graphics card three months after AMD introduced their HD 7970 graphics card. However, this time, neither AMD or NVIDIA is the cause of production issues but it is TSMC which have moved the bulk of their business to SOCs and mobile chips which have given them a huge profit over time but due to limited production, this business might soon be over as companies switch to and demand smaller nodes.
While I can confirm that AMD will have 20nm next year and 28nm in the current year, I can neither confirm or deny whether NVIDIA will have 28nm or 20nm products out this year. Unlike AMD's upcoming parts, Maxwell is a new generation of cards and a new architecture which we only saw a little glimpse in the form of the GM107 core. So whether or not NVIDIA will have something better than their competitors remains to be seen. AMD on the other hand is going for an extremely power efficienct route since the Tonga GPU is said to offer the performance on the level of R9 280X and power consumption on the level of R9 270X which is pretty impressive.