Since AMD launched EPYC 3rd generation Milan processors in mid-March, it was only a matter of time before hardware partners such as Supermicro unveiled new offerings powered by the new AMD server processors. Based on the Zen 3 architecture, AMD promises the new processors will be the highest performing x86-based server CPUs based on per-core and throughput metrics. Supermicro and AMD jointly unveiled a comprehensive range of new servers designed to meet the needs of a wide gamut of users.
Before we dive into Supermicro’s newest EPYC SKUs, let’s take a look at what separates Milan from its predecessors, codenamed Naples (Zen) and Rome (Zen 2). Compared to Rome, Milan offers twice the L3 cache capacity that’s directly accessible per core and 19 percent higher instructions per clock cycle. On the security side, Milan processors support Secure Nested Paging (SEV-SNP) and AMD Shadow Stack. In total, AMD has 15 models planned, starting with the 8-core 72F3 and going up to the 64-core 7763 and 7713/P. AMD states the 7003 series is designed with a focus on value and security for enterprise, cloud, and high-performance compute (HPC) users.
AMD Corporate Vice President John Morris notes that AMD has seen a business increase since the advent of Work from Home, and the new Milan architecture is ready to meet the increased needs of companies moving to virtual desktop infrastructure (VDI) and HyperConverged Infrastructure (HCI) to support their remote workforces. Going into the new generation, Supermicro boasts the broadest EPYC portfolio with a number of models already available. Their range of SKUs can be broken down into four segments that they actively target: GPU-oriented, density optimized, high-performance storage servers, and general-purpose servers and workstations.
GPU Optimized Servers
To meet the needs of artificial intelligence/machine learning (AI/ML), research, and graphics workload users, Supermicro has released several new products that bring together third-generation EPYC and the latest server GPUs. With the upgrade to Milan, the latest generation of GPU-optimized servers is fully PCI-E gen-4 and can support eight GPUs without any additional PCI-E switch chips needed. Both Nvidia and AMD GPUs are fully supported with the new EPYC Milan hardware.
When Nvidia GPUs are installed with both NVLink and NVSwitch enabled, each Nvidia GPU can communicate at transfer rates up to 600 GB/s, twice as fast as the previous generation. These powerful GPU-oriented systems are perfect for AI training workloads, simulation modeling, and also post-production work for films and motion pictures.
With their density optimized multi-node servers, such as their 8U SuperBlade, SuperMicro has updated their offerings to utilize the new EPYC 3rd generation processors, which promises 30% performance gains over Rome. As blade servers share a single chassis and power supply, they can be a better value for organizations that want high compute power at lower cost and with space savings.
Their new half-rack server blades can support up to two GPUs per blade for a total of forty GPUs in a single 8U enclosure. Also available with front-facing I/O for easy servicing, Supermicro’s latest blades also support hot-swapping of NVMe solid-state drives. Their latest Milan-powered blade servers were designed for cloud service providers, high-performance compute, and big data analytics.
Supermicro’s flagship 1U server, the Ultra brings all the benefits of EPYC Milan to a small form factor. The Ultra supports a large memory capacity, with 2 DIMMs per channel for a maximum of 32 DIMM slots. This brings the maximum memory supported up to 8 TB, which is double that of most 1U servers. On the storage side, the Ultra 1U server features 12 drive bays versus the previous generation’s 10. Thanks to EPYC Milan’s 128 PCI-E lanes, u.3 NVMe throughput can be effectively doubled with PCI-E gen 4 SSDs. As an additional benefit, the Ultra needs only air cooling with no modifications when only one processor is used.
The powerful computing power coupled with the small form factor makes the Ultra a great candidate for virtualization, content delivery, media streaming, and enterprise applications. It boasts the most memory and I/O capacity found in a 1U server, making it a great choice for datacenters were space is already at a premium.
Supermicro’s server platform line for data centers has been updated with support for EPYC 3rd generation processors. Designed for scalability in cloud storage data centers, the CloudDC can be found in both 1U and 2U profiles. Designed around a single CPU, 2U CloudDC servers feature configurable PCI-E gen 4 slots and support for two double-width GPUs or four single-width GPUs. It can also be configured to support up to 12 u.3 gen 4 NVMe drives. In the back, the CloudDC features dual advanced I/O module (AIOM) slots, which allows for additional network connectivity options without impacting the number of PCI-E slots.
In the data center, the CloudDC is a versatile device that can be configured for storage, which allows it to be ideal for real-time customer interactions. Tasks related to finance, commerce, and big data analysis can be handled with ease. When configured for GPUs, the CloudDC can be used for deep learning and AI tasks. As a single processor system, businesses will see lower power consumption compared to dual processor systems.