English
Supermicro rolled out its new H15 server lineup built around 6th Gen AMD EPYC 9006 CPUs, optimized for AMD Instinct GPUs and Pensando networking. The brand also expanded its AMD GPU portfolio with new PCIe servers for Instinct MI350P, alongside the previously unveiled AMD Helios rack-scale platform powered by MI455X GPUs. The launch covers CPU hardware, GPU/networking gear and integrated rack systems, targeting agentic AI, cloud, enterprise and HPC workloads.
H15 Architecture & CPU Platform
Built on Supermicro’s DCBBS modular framework, the H15 standardizes rack-scale deployment across all servers instead of custom-designing each chassis. It supports 6th Gen AMD EPYC 9006 processors with up to 256 cores and 512 threads per socket, delivering 1.7x generational performance alongside expanded memory and I/O bandwidth. Supermicro’s core value: extra cores and memory bandwidth per socket enable more AI agent or enterprise workloads per node without exceeding power caps—critical as rack power supply, not compute density, often limits data center capacity.
Six H15 System Categories
Supermicro splits the H15 range into six purpose-built server models:
1,Hyper: Flagship dual-socket unit for enterprise apps, AI inference, virtualization and cloud workloads, with cooling for top-tier EPYC chips.
2,CloudDC: Single/dual-socket variants compliant with OCP DC-MHS standards, compatible with third-party OCP hardware instead of locked Supermicro racks.
3,GrandTwin: 2U four-node high-density platform for scale-out tasks including object storage, virtualization, cloud services and HPC.FlexTwin: Compact 1U two-node dual-CPU liquid-cooled servers for hyperscale cloud environments prioritizing rack density and power efficiency.
4,Petascale Storage: 1U/2U all-flash storage appliances supporting up to 4.8PB each, built on software-defined storage for AI data lakes, analytics and HPC storage (no fixed hardware RAID).
5,SuperBlade (8U 10-node): Rack-scale blade servers with air/liquid cooling, supporting single/dual-socket blades for mixed CPU-GPU HPC, agentic AI and enterprise deployments without full dedicated GPU racks.
New PCIe GPU Servers for AMD Instinct MI350P
Supermicro launched two 5U air-cooled PCIe GPU servers (AS-5126GS-TNRT, AS-5126GS-TNRT2) housing up to 10 MI350P GPUs per chassis, compatible with existing standard data center cooling and power infrastructure and eliminating liquid cooling requirements. Each MI350P carries 144GB HBM3e memory and supports low-precision AI formats, ideal for training and inference where memory capacity, not raw compute, is the main bottleneck.
Pensando Pollara 400 Open Ethernet Networking
These MI350P servers integrate AMD Pensando Pollara 400 AI NICs, open Ethernet adapters handling front-end, storage and scale-out cluster traffic. Unlike closed proprietary interconnects limited to one GPU vendor’s ecosystem, standard Ethernet enables linear scaling from single servers to multi-rack clusters with no vendor lock-in—a key competitive advantage highlighted by Supermicro and AMD.
Supermicro AMD Helios: 72-GPU Liquid-Cooled Rack Platform
Co-developed with AMD, the liquid-cooled Helios rack holds 72 Instinct MI455X GPUs, paired with 6th Gen EPYC CPUs, Pensando networking and AMD ROCm software. It serves large-scale frontier model training and high-throughput inference, scaling from standalone racks to massive multi-rack AI clusters. The Helios and MI350P PCIe lines serve distinct use cases: MI350P air-cooled servers suit enterprises expanding GPU workloads on existing facilities, while Helios targets operators building large dedicated liquid-cooled AI clusters.
Executive Comments
Supermicro CEO Charles Liang stated the H15 launch centers on high-performance, scalable, power-efficient infrastructure amid rising agentic AI adoption. He cited DCBBS modular design as the foundation for flexible rack deployment, plus Supermicro’s global support network and U.S.-based supply chain to streamline AI infrastructure rollouts.
AMD Compute & Enterprise AI SVP Dan McNamara noted scaling agentic AI demands balanced performance, efficiency and flexible deployment. Pairing AMD’s full stack (EPYC CPUs, Instinct GPUs, Pensando networking) with Supermicro’s modular server and rack hardware accelerates AI infrastructure rollouts while boosting resource utilization, energy efficiency and total cost of ownership.
Compression Note
All technical parameters, product models, hardware specs, target workloads and executive viewpoints are fully preserved; redundant descriptive phrases, repetitive modifiers and wordy clause structures are trimmed to cut roughly 10% of original word count with zero information loss.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!