HPE ProLiant Gen13 debuts with 6th Gen AMD EPYC

The new lineup introduces four server designs featuring up to 256 cores per socket, MRDIMM memory support, and PCIe Gen6 links to handle agentic AI pipelines and local inference.

Hewlett Packard Enterprise (HPE) and AMD unveiled the first four servers in the HPE ProLiant Gen13 family on October 7, 2026. Powered by 6th Gen AMD EPYC processors (the 9006 series, codenamed “Venice”, on socket SP7), the machines combine dense x86 compute with high-speed memory and bus interfaces engineered to keep AI inference models fed inside corporate data centers.

Four HPE ProLiant Gen13 models for data centers

The initial rollout divides into two standard rackmount chassis and two high-density modules built to Open Compute Project (OCP) specifications.

For standard enterprise racks, the ProLiant DL525 Gen13 offers a 1U, single-socket design with up to 256 cores. It provides 16 memory channels compatible with DDR5 (up to 8,000 MT/s, 4 TB maximum) or MRDIMM modules (up to 12,800 MT/s, 2 TB maximum), alongside PCIe Gen6 lanes and front-bay space for two double-wide or four single-wide accelerator cards.

The ProLiant DL585a Gen13 handles heavy compute in a 2U enclosure. It houses two EPYC processors supporting up to 256 cores and 600W thermal envelopes per socket. The board includes 32 memory slots, PCIe Gen6 connectivity, direct liquid cooling from the factory, and capacity for up to eight double-wide GPUs or 16 single-wide cards.

For hyperscalers and clustered cloud deployments, HPE presented two OCP-aligned nodes: the ProLiant XD245 (a 1OU chassis housing four nodes and up to eight liquid-cooled EPYC CPUs) and the ProLiant XD285 for distributed inference.

Relieving the host bottleneck in agentic AI

The architectural choices target an operational bottleneck that has shifted over the past year. In agentic workflows, autonomous software agents coordinate multiple tool calls and query local language models in rapid succession. The limiting factor in those setups is rarely raw matrix math on the GPU; it is the latency and bandwidth of moving context data between system memory and accelerators. Combining PCIe Gen6 with 12,800 MT/s MRDIMM memory seeks to prevent GPUs from idling while waiting on host transfers.

Krista Satterthwaite, senior vice president and general manager of Compute at HPE, stated that enterprise teams need to convert agentic AI investments into daily operational returns without turning infrastructure management into an unmanageable chore. Dan McNamara, senior vice president and general manager of Compute and Enterprise AI at AMD, emphasized that coupling 6th Gen EPYC processors with Gen13 platforms provides the density required to modernize business-critical workloads as automated AI routines take hold.

Remote management via iLO 8 and rollout schedule

The ProLiant Gen13 platform also brings the debut of HPE iLO 8 (Integrated Lights-Out 8). The updated out-of-band management silicon incorporates multi-party authorization, post-quantum cryptography, and ties into Compute Ops Management software to track predictive telemetry and catch hardware anomalies before nodes drop offline.

According to HPE’s official rollout schedule, the single-socket DL525 Gen13 will ship late in 2026. The dual-socket DL585a Gen13 and the OCP-standard XD245 and XD285 nodes will reach customers throughout 2027. Neither company has released SKU-level pricing or clock frequency tables for the EPYC 9006 processor lineup.