electronics-journal.com
28
'26
Written on Modified on
Supermicro Introduces H15 Servers with 6th Gen AMD EPYC
The modular infrastructure supports AI workloads, cloud environments, and high-performance computing using advanced processors.
www.supermicro.com
Supermicro has announced its next-generation H15 server portfolio, incorporating 6th Gen AMD EPYC 9006 Series CPUs, AMD Instinct GPUs, and AMD Pensando networking built on Data Center Building Block Solutions (DCBBS) architecture. The systems support up to 256 cores and 512 threads per processor, enabling data centers to execute concurrent artificial intelligence (AI) agents and enterprise applications within existing power envelopes. According to published SPECInt Rate 2017 results, the AMD EPYC 9006 Series delivers a 1.7x generational CPU performance improvement, accompanied by twice the PCIe bandwidth and 2.6x higher memory bandwidth.
Portfolio Configurations
The H15 portfolio includes purpose-built systems tailored to specific operational demands:
- Hyper: A dual-socket flagship platform engineered for enterprise applications, AI inference, virtualization, and cloud workloads.
- CloudDC: Single- or dual-socket servers built on the Open Compute Project (OCP) Data Center Modular Hardware System (DC-MHS) specification.
- GrandTwin: A high-density 2U, four-node architecture designed for scale-out object storage, virtualization, and cloud services.
- FlexTwin: A liquid-cooled 1OU, two-node dual-CPU system maximizing compute density, capable of housing up to 168 CPUs and 43,008 cores in a 48U rack.
- Petascale Storage: High-capacity 1U and 2U all-flash platforms supporting up to 4.8 PB per system for AI data lakes and software-defined storage.
- SuperBlade: An 8U architecture supporting single- and dual-socket air- or liquid-cooled configurations for HPC and enterprise workloads.
AI Integration and Acceleration
To support intensive accelerator configurations, the AS-5126GS-TNRT and AS-5126GS-TNRT2 5U air-cooled servers accommodate up to ten AMD Instinct MI350P GPUs featuring 144GB of HBM3e memory. Network connectivity is driven via the AMD Pensando Pollara 400 AI NIC over standard Ethernet infrastructure. For expansive rack-scale AI architectures, the liquid-cooled Supermicro AMD Helios Platform integrates 72 AMD Instinct MI455X GPUs, 6th Gen EPYC processors, Pensando networking, and the AMD ROCm software stack. Dan McNamara, Senior Vice President and General Manager of Compute and Enterprise AI at AMD, noted that combining these processors, GPUs, and networking tools with modular server designs allows enterprises to deploy AI infrastructure faster while improving hardware utilization and lowering the total cost of ownership.
Additional Context
This section details technical specifications and competitive benchmarking not included in the original product announcement.
The launch of the 6th Gen AMD EPYC processors (codenamed "Turin," based on the Zen 5 architecture) positions AMD to compete directly with Intel's newly released Xeon 6 processors (such as the Granite Rapids and Sierra Forest lines) in the high-density datacenter market. As AI workloads drive thermal design power (TDP) limits past 500W per CPU and 1000W per GPU, Tier-1 Original Equipment Manufacturers (OEMs) like Supermicro, Dell Technologies, and Hewlett Packard Enterprise (HPE) are increasingly standardizing direct-to-chip (D2C) liquid cooling solutions to maintain acceptable Power Usage Effectiveness (PUE) ratings. Furthermore, Supermicro's integration of AMD Instinct MI350/MI455X GPUs and the ROCm open software ecosystem provides hyperscalers with a viable, open-standard alternative to Nvidia-dominated AI clusters, leveraging standard Ethernet (aligned with the Ultra Ethernet Consortium) to avoid proprietary vendor lock-in for backend networking.
Edited by Lekshman Ramdas, Induportals editor – adapted by AI.
To support intensive accelerator configurations, the AS-5126GS-TNRT and AS-5126GS-TNRT2 5U air-cooled servers accommodate up to ten AMD Instinct MI350P GPUs featuring 144GB of HBM3e memory. Network connectivity is driven via the AMD Pensando Pollara 400 AI NIC over standard Ethernet infrastructure. For expansive rack-scale AI architectures, the liquid-cooled Supermicro AMD Helios Platform integrates 72 AMD Instinct MI455X GPUs, 6th Gen EPYC processors, Pensando networking, and the AMD ROCm software stack. Dan McNamara, Senior Vice President and General Manager of Compute and Enterprise AI at AMD, noted that combining these processors, GPUs, and networking tools with modular server designs allows enterprises to deploy AI infrastructure faster while improving hardware utilization and lowering the total cost of ownership.
Additional Context
This section details technical specifications and competitive benchmarking not included in the original product announcement.
The launch of the 6th Gen AMD EPYC processors (codenamed "Turin," based on the Zen 5 architecture) positions AMD to compete directly with Intel's newly released Xeon 6 processors (such as the Granite Rapids and Sierra Forest lines) in the high-density datacenter market. As AI workloads drive thermal design power (TDP) limits past 500W per CPU and 1000W per GPU, Tier-1 Original Equipment Manufacturers (OEMs) like Supermicro, Dell Technologies, and Hewlett Packard Enterprise (HPE) are increasingly standardizing direct-to-chip (D2C) liquid cooling solutions to maintain acceptable Power Usage Effectiveness (PUE) ratings. Furthermore, Supermicro's integration of AMD Instinct MI350/MI455X GPUs and the ROCm open software ecosystem provides hyperscalers with a viable, open-standard alternative to Nvidia-dominated AI clusters, leveraging standard Ethernet (aligned with the Ultra Ethernet Consortium) to avoid proprietary vendor lock-in for backend networking.
Edited by Lekshman Ramdas, Induportals editor – adapted by AI.
