The High-performance computing (HPC) market is continuously gaining strong momentum, with research finding that the global market is expected to grow massively for applications requiring high data computation and increasing analysis levels. These advanced applications include high-frequency trading, autonomous vehicles, genomics-based precision medicine, computer-aided design and simulation, deep learning, and more. It refers to using powerful computing systems to quickly process massive influxes of data and solve complex problems largely fueled by advances in artificial intelligence (AI) technology. Organizations across industries are embracing AI to drive innovation and unlock new revenue streams. This AI imperative demands computing infrastructures that can process data-intensive workloads at unprecedented speeds and scale. However, this shift also brings significant challenges. While combined with rapid advancements in data center evolution and the promise of emerging technologies like quantum AI, it's clear the compute infrastructure is driving the need for the expansion of data center investments.
Supermicro is a leader in the manufacture and design of high performance servers and storage solutions based on modular and open architecture. Many of Supermicro's servers are used for complex IT requirements and for performance and computing environments that require advanced power hungry solutions which need liquid cooling. The company recently announced the launch of new H14 generation servers, GPU-accelerated systems, and storage servers featuring the AMD EPYC 9005 Series processors and AMD Instinct MI325X GPUs. With Supermicro's tight relationship with CPU and GPU suppliers, AMD and Supermicro have long partnered on solutions that serve key customers in the digital economy. This partnership again leads to AMD support for Supermicro solutions in SuperComputing 2024 conference where Supermicro showcased its latest high-compute-density multi-node solutions optimized for high intensity HPC workloads.
Supermicro solutions powered by AMD
Supermicro's new H14 family uses the latest 5th Gen AMD EPYC processors, which is powering the high demanding and intensive enterprise and HPC workloads in the industry and enable up to 192 cores per CPU with up to 500W TDP (thermal design power). The company has designed new H14 servers, including the Hyper and the FlexTwin systems, which can accommodate the higher thermal requirements. The H14 lineups also include three systems for AI training and inference workloads supporting up to 10 GPUs, which feature the AMD EPYC 9005 Series CPU as the host processor and two that support the AMD Instinct MI325X GPU.
Supermicro and AMD are collaborating and seizing the opportunity to establish themselves as leaders in AI-driven data infrastructure. According to Charles Liang, president and CEO, Supermicro, he claimed "Supermicro's H14 servers have 2.44X faster SPECrate2017_fp_base performance using the EPYC 9005 64 core CPU as compared with Supermicro's H11 systems using the second generation EPYC 7002 Series CPUs. This significant performance improvement allows customers to make their data centers more power efficient by reducing the total data center footprint by at least two-thirds while also adding new AI processing capabilities." The new H14 Supermicro product line, based on 5th Gen AMD EPYC CPUs, supports a broad spectrum of workloads and excels at helping a business achieve its goals, which can be highlighted as highest performance x86 server processor and leadership x86 energy efficiency.
H14 server family fulfills multiple needs of modern data centers
The rapid growth of consumer AI adoption has driven many large technology companies to accelerate their shift toward large language models and other AI technologies to provide innovative solutions and remain competitive in both the public and private sectors. Different types of workloads are used to accomplish different AI tasks. Various workloads are all addressed by the Supermicro H14 servers and storage systems. These include:
High-Performance Computing (HPC) – HPC systems are used by more than university and national lab researchers. Now, more enterprises integrate HPC systems into everyday workflows to bring products to market faster and discover new vaccines and drugs. Advanced HPC systems require fast cores, large amounts of memory, and fast networking between systems.
Cloud – Designing and implementing a cloud solution requires a wide range of optimized products for different workloads, not just for environments where the price performance of the compute aspect is most important. Storage and networking are critical for a productive and cost-effective cloud data center.
Artificial Intelligence (AI) – Systems with fast CPUs and associated GPU sub-systems are required for the growing AI use cases. Supermicro H14 servers can house up to 10 GPUs in a 5U rack height and excel at AI applications, enabling faster training and inference applications. Supermicro designs servers specifically to accommodate a high number of GPUs for maximum AI application performance. In addition, the Supermicro GPU servers incorporate the latest GPUs from several vendors in various form factors.
Big-Data Analysis – As the volume of data generated everywhere explodes, the systems must access, analyze, and present structured and unstructured data to the user. These tasks require the ability to hold an increasing amount of data in memory, fast computation, and quick data communication to GPUs if needed.
Virtualization – With many enterprises utilizing virtualization technologies to get higher utilization from existing servers, the new Supermicro H14 servers, with the 5th Gen AMD EPYC processors, allow for higher-powered virtualization machines, as more cores are available and faster CPUs.
Enterprise – Typical enterprise workloads will benefit from the new Supermicro H14 systems with increased performance and reduced costs. In addition, existing workloads will execute faster, using less power than previous generations of Supermicro servers.
The spotlighting features of Supermicro's H14 portfolio of servers
Managing AI workloads in data centers can be difficult if the systems aren't ready to meet the need. Networking, processing, and scalability features must be in place for AI workloads to be functional. There are several unique strengths for Supermicor's H14 servers to achieve for responding these requirements.
Broad selection—The Supermicro H14 product line offers a wide range of choices optimized for specific workloads. These product families consist of the following product families: Hyper-enterprise server, CloudDC versatile system optimized for use in cloud data centers, GrandTwin 4-node compute platform, FlexTwin 2U 4-node performance high-density compute system, and 4U/5U/8U GPU systems. The wide range of selections satisfies the expansion of data centers and the need for advanced infrastructure.
Compute Power—The Supermicro H14 products with the AMD EPYC 9005 series processors offer top-level performance for many metrics and, when combined with high core counts, are ideal for a range of workloads. The solutions are purpose built to accelerate data center, cloud, and AI workloads driving new levels of enterprise computing performance.
Max Core Counts—Supermicro H14 servers have been designed to house AMD's most powerful and energy-intensive CPUs for high-end computing environments with up to 192 cores in a single CPU, making the H14 servers the ideal rackmount solutions for Cloud, HPC, and AI applications. Furthermore, using a 48U rack as an example, FlexTwin can support up to 96 dual processor nodes and 36,864 cores within this rack size.
Max Density—Supermicro H14 multi-node architectures leverage shared resources, including cooling and power supplies, to maximize energy efficiency, with compact node form factors that allow significantly increased compute and component densities compared to standard rackmounts. Taking the all-new H14 FlexTwin as an example, this new design is purpose-built for HPC at scale, with front-accessible nodes, flexible networking and storage, and direct-to-chip liquid cooling, providing outstanding density and optimized thermal performance.
Thermal Design— by optimizing the airflow within a system, high-performing CPUs can be used without concern for overheating. Liquid cooling increases the compute density and lowers the data center power usage effectiveness (PUE).
Deliver high-performance, liquid-cooled servers for unleashing the full potential of AI
In general, high-end server systems are also high-density, have higher performance and usually better efficiency, it also means higher power density. The GPU-dense systems designed for AI workloads has driven power demands from supporting 6-12 kilowatts per rack to 40-60, and even quickly ramping to 100-150 kilowatts per rack. Not surprisingly, moving to 500 kW per rack or even 1 MW per rack is an ongoing trend. However, considering airflow and containment are excellent methods to improve efficiency and density for now, current server solutions are quickly reaching the limits of the physics of airflow. And the next logical step is to turn to liquid cooling.
Leveraging Direct-to-chip liquid cooling technology, Supermicro removes 90% of server-generated heat in FlexTwin systems. This ability has become a strategic advantage for Supermicro, aiming to harness AI server business and maintain a competitive edge. The new H14 server family showcased from Supermicro demonstrates significant technological advancements to show its strength to enhance liquid cooling capabilities. On the other hand, Supermicro works closely with customers to architect and design rack and entire data center solutions for HPC workloads. After the design is validated with customer close involvement, Supermicro offers on-site deployment services, reducing the time-to-deployment. With a global manufacturing footprint and production facilities, Supermicro can produce a total of 5,000 racks per month, including 2,000 liquid-cooled racks, with lead times of weeks, not months.
These efforts make close partnerships with AI chipmakers allow Supermicro to gain early access to their new chips and produce their multiple server family products before competitors. This has been an important advantage and has attracted global hyperscalers to drive demand for its AI infrastructure. The pace of adoption of advanced AI use cases will certainly continue to evolve. The new server design will mix the of different types of chips deployed and their associated power consumption, as well as the balance between cloud and edge computing for AI workloads and the typical compute, storage, and network needs of AI workloads. Supermicro carved out its own niche by selling high-performance, liquid-cooled servers and quick ramping manufacturing capacity for demanding computing tasks. That made it an ideal partner for global chip makers like AMD, which supplied Supermicro with high-performance data center CPUs and GPUs to help it produce dedicated AI servers to grasp market share.
Supermicro's H14 family is powered by the 5th Gen AMD EPYC processors which enable up to 192 cores per CPU with up to 500W TDP