At its financial analyst day, held yesterday, AMD finally unveiled its long-awaited Zen architecture. As recent rumor has said, chips featuring the new core will be released at some point in 2016.
AMD touts Zen as being developed with a “totally new” core design, one that supports both SMT and a high-bandwidth / low-latency cache system. As expected, it will also support DDR4 memory, which, by the time Zen gets here, should have the early-adopter’s fee long behind it.
As seen in the slide below, Zen offers a 40% IPC boost over Excavator, and down-the-road, we can expect to see Zen+ cores come out. One thing worth noting is that AMD’s Zen-based desktop CPUs will utilize the same socket – an FX chip can be used just like an A10 APU in an AM4 socket.
On the IGP side, 2016’s GCN architecture will deliver high performance with twice the energy efficiency, and like the CPU core, will be built using a FinFET process technology. As with the current GCN architecture, the one bundled in with Zen will support DX12, Vulkan, and Mantle.
AMD’s pride and joy with Zen will be its use of high-bandwidth memory (HBM) for the GPU, as it expects to be first to market. HBM can increase the performance-per-watt by 3x, and shave power draw in half.
Here’s something even more forward-thinking to get intrigued about: AMD’s 7th-generation APUs promise “discrete” like GPU performance and full HSA performance in a socket designed for Ultrathin notebooks, called FP4.
Everything AMD has laid on the table so far looks good, but unfortunately we’ll have to wait a bit longer before we get the nitty-gritty and can really gain an understanding of what Zen could deliver.