HPE's first ProLiant Gen13 servers use AMD's 256-core Venice chips, with the first shipping in November
The 1U DL525 ships first. The eight-GPU DL585a follows in spring 2027, and two dense multi-node systems come later in the year.
İsmail Enes Ayhan / UnsplashHPE announced its first four ProLiant Gen13 servers on October 7. All four use AMD's sixth-generation EPYC processors, code-named Venice, with up to 256 cores per socket. HPE is pitching them at AI inference and agent workloads that need a lot of CPU alongside a few GPUs, as well as at general virtualization.
"These are not the big NVL72 or AMD Helios scale-up type systems," John Carter, HPE's vice president of server product management, told Data Center Knowledge.
The four models
| Model | Form factor | Main specifications | Availability |
|---|---|---|---|
| ProLiant DL525 Gen13 | 1U, one socket | Up to 256 cores, 16 memory channels, up to 4TB DDR5-8000, up to 20 E3.S NVMe drives | November 2026 |
| ProLiant DL585a Gen13 | 10U, two sockets, air-cooled | Up to 8 double-wide GPUs or 16 single-wide liquid-cooled GPUs, up to 8TB DDR5-8000 | Early 2027 (spring, per HPE) |
| ProLiant XD245 Gen13 | 2U, four nodes | Two sockets per node, direct liquid cooling only | 2027 (around early summer) |
| ProLiant XD285 Gen13 | 2U, two nodes, air-cooled | Same per-node specification as XD245 | 2027 (around early summer) |
The DL585a will take GPUs from Nvidia, AMD and Intel. Each Venice socket can draw up to 600W, so check rack power and cooling budgets before ordering.
Management and security
The servers come with HPE's iLO 8 management controller, which adds support for self-encrypting drives and more post-quantum cryptography readiness. HPE Compute Ops Management gets low-touch onboarding and near-real-time readings of power capacity.
HPE did not publish prices. HPE Financial Services is offering a 90/9 plan with no payments for 90 days, then nine months at 1% a month, StorageReview reported.
Why the CPU matters again
Analysts quoted by Data Center Knowledge say agent workloads lean on the CPU to coordinate many model calls, tools and data lookups. "That high core count and high single-threaded performance do a very good job of supporting orchestration for agentic AI," said Matt Kimball of Moor Insights & Strategy.
What IT teams should do
- If you are planning a 2027 server refresh, ask HPE for DL525 Gen13 evaluation units. It ships first.
- Check power and cooling before you order. Each Venice socket can draw up to 600W, and the XD245 needs direct liquid cooling.
- If you are buying GPUs for inference, compare the DL585a against GPU-dense systems on cost per query, not on peak specifications.