Supermicro ships H15 servers on AMD EPYC 9006

Super Micro Computer (SMCI) has started shipping its H15 generation server portfolio built on 6th Gen AMD (AMD) EPYC 9006 Series processors. The launch targets workloads including AI inference, cloud computing, high-performance computing, storage, virtualization, and enterprise applications. The new systems are designed to advance performance for cloud, HPC, and agentic AI deployments while maintaining flexibility across single-socket and dual-socket configurations.

The H15 portfolio delivers up to 33% more cores, 2× PCIe bandwidth, and 1.6 TB/s per-socket memory bandwidth compared with previous generations. The servers also provide 1.8 times higher tokens per second for frontier-model host nodes relative to prior AMD CPU generations, positioning the platform for demanding AI and HPC use cases.

Each H15 system can be configured with AMD’s sixth-generation EPYC 9006 Series processors, codenamed Venice. The headline capability is density, with up to 256 cores and 512 threads per socket. Systems are available in both single-socket and dual-socket designs, allowing buyers to tune deployments for either maximum compute per node or broader scaling across racks.

Each socket supports 16-channel DDR5-8000 memory, delivering up to 1.6 TB/s of memory bandwidth per socket. This high bandwidth is aimed at bandwidth-intensive workloads such as HPC and AI inference, where feeding large models and datasets quickly is critical. The platform is specified to operate at thermal design power levels up to 600W per socket, with designs accommodating both air and liquid cooling to manage thermals at higher performance points.

On the input/output side, the H15 platform supports PCIe 6.0 with up to 96 PCIe 6.0 lanes per socket. This expanded connectivity is intended to support high-speed networking, storage, and accelerator attachment. The portfolio is optimized to work with AMD Instinct accelerators and AMD Pensando networking technology within Supermicro’s Data Center Building Block Solutions architecture.

The H15 range spans multiple product families, including Hyper, CloudDC, GrandTwin, FlexTwin, Petascale, and SuperBlade. GrandTwin provides a 2U 2-node architecture for scale-out environments, while Petascale Storage platforms support up to 4.8 PB per system. The 6U SuperBlade system supports both single-socket and dual-socket blades in air- and liquid-cooled variants, focused on high-density and efficiency.

In parallel with the H15 server rollout, Supermicro and AMD are developing Helios and Helion rack-scale solutions for large-scale and agentic AI deployments. Both platforms are built on Supermicro’s DCBBS architecture and leverage AMD EPYC CPUs, with optimization for AMD Instinct GPUs. The goal is to provide complete end-to-end AI infrastructure options from agentic to hyperscale environments.

Helios is designed for maximum GPU performance and is powered by AMD Instinct MI455X GPUs to accelerate frontier model training and high-throughput inference. Helion focuses on compute performance and deployment flexibility using the latest 6th Gen AMD EPYC CPUs. Together, these rack-scale offerings and the H15 server portfolio are intended to give enterprise customers higher performance and density across AI, cloud, and HPC workloads.