At AMD’s annual Next Horizon event Nov. 6, in San Francisco, it announced details of its next-generation data processors. After introducing the first Zen processor in 2017, the company failed to gain commercial success. With this recent announcement, it seems AMD is ready to take on its competitors in the GPU space for the data center.
The 2018 Next Horizon event featured three major product announcements:
- A number of deals with Amazon, Microsoft, Cray and Baidu
- The launch of the 7nm AMD Radeon Instinct, based on the Vega 20 GPU, as a data center GPU accelerator
- Introducing the next-in-line in the Zen Family, Zen 2, code-named Rome
The event highlighted deals between AMD and AWS, as well as several others in the high-performance computing (HPC) community. AWS will be announcing the availability of the three machine instances based on Amazon Elastic Compute, which will use AMD Zen cores R5a, M5a and T3a. Cray also announced its deal with AMD to add EPYC processors to its Cray CS500 product line, Shasta. The integration aims to enable organizations to handle HPC workloads without having to rebuild and recompile their x86 applications. AMD’s CEO, Lisa Su, shared that sales from its first-generation Zen processors have propelled them to reach mid-single digital market share.
The biggest announcement of the event was the unveiling of the AMD Radeon Instinct M160 and M150 accelerators. AMD labels them as “the world’s first 7nm datacenter GPUs.” The GPUs are based on AMD’s Vega architecture and are intended to power HPC, deep learning, cloud computing, VDI and rendering applications. They are optimized to perform deep-learning capabilities and can successfully train complex neural networks, as well as run inference against those networks.
To process these workloads, AMD said the accelerators enable “ultra-fast” floating-point performance and feature HBM2 memory, which can reach speeds up to 1TB/s.
AMD will also be introducing the Zen 2 x86 processor, which has adopted a “chiplet” approach. According to AMD’s CTO, Mark Papermaster, AMD made the decision to leapfrog 10nm two years ago as its engineers recognized that it wasn’t something customers needed. Additionally, Papermaster noted that the constraints of physics have made new process technologies increasingly expensive and complicated to produce. AMD instead invested in Infinity Fabric technology for faster interconnects, providing the most architectural flexibility for the company.
The decision to skip 10nm has positioned AMD ahead of Intel in producing 7nm, for which Papermaster said they give credit to their long-term fab partner TSMC.
Combined Zen 2 chiplets built on TSMC’s 7nm processor will be connected via an enhanced Infinity Fabric link to a 14nm I/O core. The 14nm core will include a memory controller plus DDR DRAM interfaces, which enables equal memory access to the CPUs.
AMD also gave details about its next generation of EPYC processors, Rome. It will include up to 64 Zen 2 cores. Rome will be the first x86 7nm CPU to hit the market.
For more news on GPUs and processors, check out New CPU Increases Engineering Workstation Performance. If you want to support similar projects or create your own, check our ProjectBoard for the latest ideas, like the OMNIBoard.