Cloud servers are evolving from legacy, x86-based hardware to cloud native platforms that can support workloads that natively run in the cloud infrastructure. Ampere^® Altra^® and Ampere^® Altra^® Max aarch64-based processor cores integrated with high-bandwidth 128-PCIe Gen4 lanes and 8-channels of DDR4-3200 memory can be optimized with software at the platform level to provide flexible, scalable, and secure cloud native servers. Ecosystem partners including Foxconn, Gigabyte, Inspur, Supermicro, and Wiwynn are enabling the cloud with Ampere to provide cost-effective scalability for evolving workloads.

The movement of legacy workloads to the cloud leads to fewer, larger, more efficient datacenters. The size of these datacenters has driven cloud service providers (CSPs) to refocus on power consumption and efficiency to minimize the total cost of ownership (TCO) as well as to embrace a more sustainable infrastructure. The Ampere Altra and Ampere Altra Max processors can deliver performance superior to x86 CPUs while consuming significantly less power.

As shown in Figure 1, benchmarks for Ampere Altra Max Q128-30 compared to flagship AMD Milan EPYC 7763 chips show consistent improvements in real-world metrics for server functionality enabling cloud service providers to have 62% higher performance per core, 8% lower power and 80% higher performance per watt.

Disruption of legacy cloud business models may be seen in the recent announcement that the Oracle Cloud Infrastructure (OCI) offers Ampere Altra 80-core processor server rental at only $0.01 per core hour, with flexible sizing from 1-80 CPUs and 1-64 GB of memory per core. The OCI Ampere Altra A1 compute instance delivers predictable performance, linear scalability, security and power efficiency along with the best price-performance in the market. An entire Oracle ecosystem exists for developers to transition, build, and run aarch64-based workloads.

Enabling new cloud business may happen through the Open Compute Project (OCP). OCP was initiated by Facebook in 2011 with a mission to apply the benefits of open source and open collaboration to hardware and rapidly increase the pace of innovation in the data center. Ampere^® has been a platinum member of OCP since 2019 and remains committed to open hardware and source firmware.

Platform Flexibility for Changing Workloads

The combination of single threaded cores and consistent operating frequency allows Ampere Altra and Ampere Altra Max to deliver predictable performance. The ability to hit and maintain maximum CPU frequencies results in higher predictability along with fewer "noisy-neighbor" problems for developers in the public cloud. This leads to new platform form-factors, configurations, and innovative designs at both infrastructure and workload levels (Figure 2) that can take full advantage from the high core counts of Ampere CPUs.

Microservices-based and container-based usages and workloads benefit greatly from the increased core density of Ampere Altra and Ampere Altra Max. Cloud native applications are specifically designed to run in the modern data center environment and cover a wide range of capabilities. A rapidly evolving set of business segments with common data privacy and security requirements is termed Virtual Mobile Infrastructure (VMI), and includes education, healthcare, government, finance, live streaming and social media. IDC estimates that 80% of workloads will shift to containers/microservices by 2023.

Workload volume scaling happens both gradually over time due to natural growth in demand, and also suddenly when a huge workload must be run in parallel across multiple servers. CSPs require that infrastructure investments be "future-proofed" using flexible platforms capable of transforming into Virtual Machines (VM) in response to diverse user applications. Ampere Altra and Altra Max high core counts combined with low power consumption/core enables increased containers and microservices per CPU at a rack-level, leading to higher revenues.

As one specific example, cloud gaming on Arm-based servers like Ampere consists of a large application base of gaming titles that are native to Android OS and primarily targeted at vast-scale consumer mobiles/tablets worldwide. Ampere processors allow virtually unrestricted access to server-side resources to run ever more demanding games for streaming to any device. Game companies avoid the restrictions and compatibility issues of developing for a myriad of mobile devices.

CSPs using Ampere Altra and Ampere Altra Max CPUs for a variety of cloud native workloads involving web serving, caching, databases, analytics, and AI receive the benefit of higher core counts, lower power consumption/core, and increased VM/user density at a rack-level, all of which leads to higher revenues. Scaling from cores to servers to racks, compute requirements from the smallest fractional workloads to the largest clusters get distributed efficiently.

Robust Ecosystem

Original Design Manufacturer (ODM) support for Ampere Computing is demonstrated by the availability of >26 unique single and dual socket servers (Figure 3). Ampere works closely with ecosystem partners to develop an approved vendor list (AVL) – including memory and storage drives from Kioxia, Micron, Samsung, and SK hynix; network interface cards (NIC) from Broadcom, Intel, and Mellanox; and other acceleration (GPU/FPGA) cards from AMD and NVIDIA. ODM and OEM platforms are able to leverage the tested Ampere approved vendor list (AVL) to provide unique hardware and software solutions for customers.

Examples of Ampere platform solutions for growing cloud workloads include the following:

• Hamilton (1P) and Mt. Collins (2P) systems targeted for Android In Cloud (AIC), AI, analytics, and high-performance workloads to maximize use of eight NVIDIA Tesla T4 GPUs with excellent transcoding,
• Snow IP server optimized for Java performance at impressively low watts/core, and
• Bonnell 1P half-width server for general compute with high-performance NVMe drives and high end E1.S storage.

Partners in the Ampere ecosystem are investing engineering resources and have committed to continue investing in this cost-effective platform to grow the cloud. At the same time, Ampere is committed to open source firmware and this helps ensure that datacenter workflows run seamlessly on different platforms by making ongoing contributions to TianoCore/EDK2, LinuxBoot, OpenBMC, and OpenOCD.

Ampere has worked with its ODM/OEM partners to ensure that despite recent supply-chain disruptions, global customers have been supplied with servers throughout 2021 and robust plans are in place to have a steady supply in 2022. With a strong hardware roadmap, continuous improvement in software, and an expanding AVL to meet global market needs, the Ampere cloud-native ecosystem is setting new standards for CSPs.