Focus on Computing and Graphics
In the APU market AMD outlined plans to change the way it delivers products for different market segments. AMD reiterated on several occasions the need to fully monetise the APU concept, particularly in desktop and high-performance computing (HPC) markets. To date AMD has struggled to do this as many of its APU architectures have not scaled-up effectively to those 65W to 100W thermal envelopes. As part of AMD's new APU strategy there will be a dedicated high-performance server APU to help target the datacentre market while lower-power markets will be based off a different APU design.
AMD pledged a 'transformational memory architecture' for these high-performance APUs which can presumably overcome the system memory bottleneck that current processors face. In doing this AMD believes it can deliver 'Multi-Teraflops' for HPC and Workstation users with graphics performance that scales up effectively.
On the consumer side AMD is close to launching its 6th Generation Carrizo APUs that will arrive in desktop and notebook form-factors this quarter. These will bring improved graphics performance over Kaveri, improved power efficiency and battery life and support for the High Efficiency Video Coding (HEVC) standard that allows for smooth, low-bandwidth playback of video in resolutions up to 4K.
Carrizo is to be succeeded by AMD's 7th Generation of A-Series APUs in 2016 which will continue to increase performance-per-watt efficiency as part of AMD's 25x20 initiative. The 7th Gen APUs will have support for the AM4 desktop socket, shared by Zen-based FX CPUs, and there will be an FP4 mobile infrastructure for mobile device vendors.
AMD rounded off its APU discussions by revisiting the custom K12 APU that leverages a 64-bit ARM architecture. AMD is delivering these ARM-based APUs to the market under the Opteron A1100 series and expects availability to proliferate in the second half of 2015. The "Seattle" Opteron A1100 series processors are focused on high-performance server markets that require energy efficiency and hardware-level security features.
GPUs
In the graphics segment AMD has similarly adventurous plans, although, like with Zen, these will have to wait until 2016 to be delivered in full. For 2015 AMD says we can expect to see incremental enhancements to its GCN architecture as soon as this quarter. These enhancements will be centred on the new High-Bandwidth Memory (HBM) which AMD claims will deliver a 3x performance-per-watt improvement over GDDR5 memory. The ability to stack VRAM dies vertically provides AMD with the ability to make smaller form-factor GPUs to target new markets.
Additionally, AMD says it will be continuing to pursue compatibility for this year's GPUs with leading-edge API projects like DirectX 12, Mantle and the recently announced Vulkan. Another focus of the company is tuning its GPUs for virtual reality devices that are part of the strategy to target the immersive experiences and gaming markets. AMD noted on several occasions that its FreeSync technology would be making its way to mobile GPUs, of which AMD plans to make the new M300 series of Mobility Radeon GPUs available to OEMs imminently.
In 2016 AMD will launch new graphics products based on FinFET technology though the company didn't specify exactly which process node size it would be using - 20nm, 16nm or 14nm. These new FinFET-based GPUs deliver 2x performance-per-watt and will implement second-generation HBM technology.
AMD has divulged more details during this FAD than on previous ones. The strategy seems coherent at first glance, but as with everything, the proof will come when products are on the shelves. AMD is fighting on multiple fronts with relatively limited resources, we note, so it will be interesting to see how things play out into 2016 and beyond.
