The View from Our Corner of the Street
November is shaping up to be quite a month! In addition to 2nd Generation AMD Ryzen Processors winning multiple industry awards including “CPU of the Year” for the AMD Ryzen 7 2700X and “Value CPU of the Year” for the AMD Ryzen 5 2600X in the prestigious Trusted Reviews Awards 2018, we also made several game-changing announcements at our AMD Next Horizon event on November 6th.
Dr. Lisa Su, President and CEO, opened Next Horizon with an overview of the amazing progress we have made in the datacenter including the immediate availability of three EPYC CPU based EC2 instances on AWS. This is an important announcement from an adoption perspective, but also serves as validation of the TCO advantages that EPYC and the Zen architecture bring to the most cutting-edge datacenters.
Mark Papermaster, SVP CTO, discussed our Zen core roadmap, with Zen/Zen+ shipping today, Zen 2 sampling, Zen 3 development on track, and Zen 4 in design. We are now on the leading edge of process technology with 7nm GPUs and CPUs. But advanced process technology is only one aspect of what makes Zen 2 great. Zen 2 is also a completely new architecture, which takes advantage of Infinity Fabric 2 to move 7nm compute “chiplet” dies to the perimeter of a 14nm I/O die, improving performance, power consumption, and latency.
David Wang, SVP Engineering Radeon Technologies Group, presented our view of the evolving use cases of datacenter GPUs, including cloud gaming, virtual desktops, machine learning, and high-performance compute. But the highlight of his presentation was our announcement of the world’s first 7nm GPU, the AMD Radeon Instinct MI60. MI60 is a huge leap from our MI25 card, offering up to 2x more transistor density, more than 1.25x higher performance per watt, and 50% lower power consumption at the same frequency. It is in fact, it is the world’s fastest FP64 and FP32 PCIe capable GPU and the first to support PCIe 4.0.
The application of Zen technology was then exhibited by Forrest Norrod, SVP GM Datacenter and Embedded Solutions Group, who discussed EPYC’s unrivaled scale and efficiency for AI, Big Data, and Cloud Computing services, and significant TCO advantages across datacenter architectures. In fact, the EPYC single socket format is so disruptive it can reduce TCO up to 45% on a cost/year/VM basis.
Not to be outdone, Lisa returned to stage for one more thing…EPYC 2 “Rome.” EPYC 2 is based on the Zen 2 architecture discussed by Mark, with up to 64 physical compute cores, offering 2x the performance per socket and 4x the floating point per socket as EPYC 1. In her mic drop moment, Lisa demonstrated one EPYC 2 “Rome” chip beating two Skylake 8180M Platinum chips (intel’s best in class platform) in a C-Ray Industry Standard benchmark. You can watch the whole event here.