AMD is going for the jugular when it comes to comparisons with Intel’s Xeon family, providing up to 128 PCI Express 3.0 lanes, which Su says “allows you to connect more GPUs directly to the CPU than any other solution in the industry."
As we’ve previously reported, EPYC scales to 32 cores/64 threads and supports up to 8-channel DDR4 memory (16 DIMMs per CPU, up to 4TB total memory support). As important as Zen architecture is to AMD, the company doesn’t want to discount the importance of Infinity Fabric, which in essence is an extension of HyperTransport. Based on Coherent HyperTransport, with some added enhancements thrown in for good measure, Infinity Fabric is what Zen and Vega ICs will use to communicate with one another.
Infinity Fabric is rumored to scale from as “little” as 30 GBps on up to 512 GBps when used in conjunction with Vega-based GPUs and AI accelerators. It's the back-end communications link, if you will, that allows AMD's advantage in front-end PCI Express expansion and offer all those extra PCIe 3.0 lanes.
AMD also noted that the Zen architecture has a roadmap to 7nm (nanometer) semiconductor fab process technology and beyond, targeting 2018 through 2020, roughly, for the introduction of Zen 2 and Zen 3.