HPE (NYSE: HPE) announced the ProLiant XD245 Server on October 7, 2026, as one of four systems in the ProLiant Gen13 family. The document on TechPowerUp's page names no price for the XD245. It is a one-Open Rack Unit, four-node, direct liquid-cooled platform that accepts up to eight 6th Gen AMD EPYC Server processors. The availability window is 2027, with no quarter specified. The other three systems in the announcement are the DL525 Gen13, the DL585a Gen13, and the XD285, but the XD245 is the only one that combines four nodes, liquid cooling, and OCP compliance in a single rack unit.

The document describes a single configuration for the XD245: four nodes in one rack unit, with a maximum of eight EPYC processors across the unit. There is no lower-tier configuration named, no entry-level SKU, no price bracket. The stretch is the full eight-EPYC build, which is also the only build described. A buyer cannot tell from this page what a two-EPYC or four-EPYC configuration would cost, or whether those configurations exist at all. The starting price buys the same thing the headline number describes, because there is only one number. This is unusual for a server announcement, where vendors typically list a base configuration and a maximum configuration with separate pricing.

The XD245 is OCP-compliant, meaning it follows the Open Compute Project's rack-level specifications for mechanical, electrical, and thermal interfaces. HPE pairs the four-node design with shared power and shared cooling, so the unit draws from a common power distribution and a common liquid loop rather than four independent cooling chains. The document does not state the power draw per node or per unit, nor the coolant flow rate, nor the thermal design power of the EPYC sockets in this configuration. The shared-loop design implies that a single coolant failure affects all four nodes simultaneously, a risk profile that differs from four independent cooling paths.

HPE positions the XD245 for CPU-intensive workloads at the rack level: computer-aided engineering, weather and climate modeling, scientific research, and financial modeling. These are workloads where the bottleneck is floating-point throughput and memory bandwidth rather than GPU tensor cores. The document does not name a specific EPYC SKU, a clock frequency, a memory capacity per node, or a TDP figure. "6th Gen AMD EPYC" is the only processor identifier in the text, and it carries no model number, no core count, and no process node. The memory type is not specified for the XD245 specifically, though the broader Gen13 family references DDR5 and MRDIMM support.

The security layer is the most detailed section of the release. HPE iLO 8 provides what the document calls "automatic protection, infrastructure trust, and post-quantum readiness." Three features are labeled industry-first: Auto-SED, which encrypts drives from first boot without external key managers or manual activation; multi-person authorization, which requires two users to approve key provisioning or secure erase; and expanded post-quantum cryptography readiness that extends beyond firmware to server communication paths. A fourth feature, enterprise-grade component verification built on HPE's Silicon Root of Trust, validates component authenticity as parts are added, replaced, or modified throughout the server lifecycle. The document does not name the specific PQC algorithm or the cryptographic standard it targets.

The reliability claims are specific but self-reported. Each ProLiant generation undergoes more than 1,000 tests and 10,000 hours of validation, the document states. HPE says 81% of ProLiant customers cite reliable performance as a key differentiator from other server vendors, and that HPE Tech Care Service delivers up to 97% less unplanned downtime. These are vendor metrics, not independent measurements. The document does not name the test methodology, the sample size for the 81% figure, or the conditions under which the 97% downtime reduction was measured. The 90/9 Advantage financing program is mentioned: no payments for 90 days, followed by nine months at 1% per month.

The XD285, the other OCP system in the Gen13 announcement, uses a two-Open Rack Unit, two-node, air-cooled design. It is the air-cooled sibling to the XD245's liquid-cooled architecture, extending the same compute platform to data centers that cannot support direct liquid cooling. The XD285 occupies twice the rack space for half the node count, which is the trade-off the air-cooled design imposes.

The constraint here is not performance. The constraint is information. A buyer evaluating the XD245 against a competing four-node liquid-cooled platform has no clock speed, no memory bandwidth figure, no power draw, no per-node throughput number, and no price. The document proves that HPE has a four-node, eight-EPYC, liquid-cooled OCP box shipping in 2027. It does not prove that the box is faster, cheaper, or more power-efficient than the alternative. The security features are real and specific, but they are table stakes for enterprise buyers by 2027; the Auto-SED and MPA capabilities will likely appear in competing platforms within a generation. What the document does prove is that HPE is committing to a shared-liquid-loop, four-node density that forces the cooling infrastructure to be designed around the rack rather than the node. That is an architectural commitment, not a marketing one. The fab implication is that the EPYC die must be thermally validated for a shared-loop environment at four-node density, which is a different thermal envelope than a single-socket air-cooled chassis.

No page in the packet measured a single benchmark, named a specific EPYC SKU, stated a price, or provided a die shot. The document is a press release relayed through TechPowerUp, and it contains no independent verification of any performance claim. The 6th Gen AMD EPYC processor is referenced but never identified by model number, core count, or process node. The shared liquid loop's failure mode, the per-node power budget, and the memory bandwidth per socket are all absent from the text.

The document proves HPE is committing to a shared-liquid-loop, four-node OCP density that forces cooling infrastructure to be designed around the rack rather than the node. It does not prove the box is faster, cheaper, or more power-efficient than a competing four-node platform, because no performance figure, price, or power number is stated.

No benchmark, no EPYC SKU, no price, no power draw, no memory bandwidth, and no die shot appear in any document in the packet.

The XD245's shared liquid loop at four-node density is an architectural commitment that changes the thermal envelope for the EPYC die relative to air-cooled single-socket designs. Without a stated TDP, memory bandwidth, or per-node power budget, the thermal and power implications of the shared loop cannot be evaluated against competing platforms.