Supermicro Launches Servers and GPU Accelerated Systems with AMD EPYC 9005 Series CPUs and Instinct MI325X GPUs for AI Ready Data Centers

Supermicro, Inc. launches a series of servers, GPU-accelerated systems, and storage servers featuring the AMD EPYC 9005 Series processors and AMD Instinct MI325X GPUs.

The H14 product line represents one of a most extensive server families in the industry, including the firm‘s Hyper systems, the Twin multi-node servers, and AI inferencing GPU systems, all available with air or liquid cooling options. The ‘Zen5’ processor core architecture implements full data path AVX-512 vector instructions for CPU-based AI inference and provides 17% better instructions/cycle (IPC) than the previous 4^th Gen EPYC processor, enabling more performance/core.

The company’s H14 family uses the 5^th Gen AMD EPYC processors which enable up to 192 cores/ CPU with up to 500W TDP (thermal design power). The firm has designed H14 systems including the Hyper and FlexTwin systems which can accommodate the higher thermal requirements. The H14 family also includes 3 systems for AI training and inference workloads supporting up to 10 GPUs which feature the EPYC 9005 Series CPU as the host processor and 2 which support the Instinct MI325X GPU.

“Supermicro’s H14 servers have 2.44X faster SPECrate 2017_fp_base performance ⁽¹⁾ using the EPYC 9005 64 core CPU as compared with Supermicro’s H11 systems using the second generation EPYC 7002 Series CPUs,” said Charles Liang, president and CEO, Supermicro. “This significant performance improvement allows customers to make their data centers more power efficient by reducing the total data center footprint by at least 2-thirds ⁽²⁾ while also adding new AI processing capabilities. The H14 server family provides the highest performance, density, and power efficiency available through Supermicro’s liquid and air-cooling options, wide selection of system designs, and proven Building Block Solutions.“

“Supermicro’s ‘Building Block Solutions’ has enabled it to consistently deliver time-to-market solutions powered by AMD across a variety of compelling system designs,” said Forrest Norrod, EVP and GM, data center solutions group, AMD (Advanced Micro Devices Inc.). “Our collaboration with Supermicro, along with their in-house engineering design and manufacturing capabilities worldwide combined with their rack-scale integration capability for both air and liquid-cooled systems enables customers of any scale to generate time-to-value from AMD EPYC CPUs and Instinct GPUs.“

H14 server lineup consists of following product families:

• Hyper – Enterprise server supports maximum performance from 2xEPYC 9005 CPUs with up to the maximum 192 cores/CPU at 500W and up to 9TB memory in 24 DIMM slots.  With either a 1U chassis with up to 12x2.5″ NVMe/SATA bays or a 2U chassis with up to 24x2.5″ NVMe/SATA bays, the Hyper’s advanced cooling design accommodates the highest performance CPUs for use in demanding AI inference, enterprise or cloud workloads.

• CloudDC – This versatile system is optimized for use in cloud data centers and features a single EPYC 9005 CPU with up to 12x2.5″ NVMe/SATA drive bays in a 1U chassis. It’s designed using the OCP (Open Compute Platform) DC-MHS specification (Data Center Modular Hardware System) which ensures compatibility with OCP standards.

• GrandTwin – The GrandTwin is a 4-node compute platform using single EPYC 9005 CPU in a dense 2U form-factor. This system is often used in multi-server cluster applications such as object storage, virtualization, and HPC applications.

• FlexTwin – The FlexTwin is a 2U 4-node performance high-density compute system with dual EPYC 9005 CPUs/node. The advanced liquid-cooling provides high power efficiency and allows the operation of the performance EPYC 9005 CPUs for use in HPC, EDA and other demanding workloads.

• 5U GPU system – EPYC CPU-based 5U PCIe GPU system supports up to 10 double-width accelerators for design and visualization applications.
• 4U GPU system (liquid cooled) – This highly dense EPYC CPU-based 8-way accelerators platform supports OAM accelerators in a compact 4U form-factor using advanced liquid-cooling. It is designed for high-performance AI and HPC applications.

• 8U GPU system – This 8-way accelerator system uses the Instinct MI325X GPU along with the EPYC 9005 CPU for large-scale LLM AI training. The 8U chassis allows deployment in any air-cooled data center.

The company’s H14 products with EPYC CPUs are available for customer testing through Supermicro’s JumpStart program. The firm will present the H14 solutions at the AMD Advancing AI Day, October 10, 2024, at San Francisco Moscone Center, CA.

⁽¹⁾ SPECrate 2017_fp_base of 485 using the Supermicro A+ Server 2023US-TR4 and 2 AMD EPYC 7702 64 core CPUs (retrieved October 2, 2024) compared with SPECrate 2017_fp_base of 1670 using the Supermicro AS-2126HS-TN and 2 AMD EPYC 9555 64 core CPUs. Results of the AS-2126HS-TN with the 9555 CPU to be published on SPEC (Standard Performance Evaluation Corporation) website on October 10, 2024.
⁽²⁾ Based on the SPECrate 2017_fp_base comparison in footnote (1), reduction in number of systems using the AS-2126HS-TN with the 9555 CPU vs. 2023US-TR4 with the 7702 CPU is 70.9%.