Park Place updates Entuity to monitor AI GPUs & cooling

Park Place Technologies has updated Entuity Software to monitor AI infrastructure across AMD, Intel and NVIDIA from a single platform. The release adds oversight of both GPUs and liquid cooling systems.

Entuity V23.0 P03 is aimed at enterprise IT teams managing AI environments that combine accelerators, servers, interconnect fabrics and cooling equipment. It is designed to give operations teams a consolidated view of hardware health and performance, with a focus on issues that could affect uptime before a full outage occurs.

The update extends monitoring beyond operating-system-level visibility by letting users connect to GPUs through an IP address defined within an operating system. This provides more detailed infrastructure data on accelerators, including heat, power draw, load and health status.

It also adds support for cooling infrastructure used in AI deployments, including direct-to-chip liquid cooling systems, which are becoming more common as newer AI chips generate more heat. In these environments, changes in cooling conditions can force GPUs to throttle performance or go offline.

The release supports GPU monitoring across AMD, Intel and NVIDIA, while also tracking the systems linked to them. These include GPU links and fabric switches, as well as cooling metrics such as pump flow, coolant temperature, pressure, reservoir level, valve position and leak detection.

According to Gartner research cited by Park Place, a rack of AI servers can draw 30kW to 120kW or more, compared with 5kW to 15kW for a standard server rack. That higher power demand raises heat loads and makes cooling a central issue for organisations deploying AI infrastructure on premises.

Fault detection

A central part of the update is earlier identification of faults that do not immediately trigger a standard availability alert. These so-called grey failures can reduce performance over time without causing a clear outage, making them harder for IT teams to detect with conventional monitoring tools.

The software also tracks memory health indicators, ECC errors and other telemetry intended to highlight hardware degradation before it leads to workload disruption. This can help users identify emerging GPU problems before a device goes offline.

For thermal management, the platform correlates GPU temperatures, power consumption and cooling data so teams can spot conditions that may lead to throttling, instability or downtime. It also includes visibility into GPU partitions, accelerator instances, memory usage and utilisation data to show where AI resources may be underused.

Network links between accelerators are another area of focus. Entuity V23.0 P03 includes switch-level visibility into the interconnect fabrics that connect GPUs, helping users identify link degradation before a failure interrupts a cluster.

"Customers can start at the server level, pull every accelerator in their environment into one consolidated view, then drill into individual GPUs to inspect load, heat, power draw, and health status in real time," said Daniel Nelson, Chief Product Officer, Park Place Technologies.

"It's a level of visibility that no other single vendor can offer," Nelson said.

Broader context

The update reflects a wider shift in enterprise infrastructure management as organisations move AI workloads on premises and confront the operational demands of denser, more power-hungry systems. Monitoring tools built for conventional servers and networks have often offered less detail on accelerators and their supporting cooling environments.

The release includes support for in-band and out-of-band Redfish through iDRAC and iLO, with in-band taking priority to avoid duplicate GPU counts. The software monitors a range of conditions, including corrected and uncorrected ECC events, link CRC, replay and recovery, fabric fatal errors, enforced power limits and PCIe throughput.

It also includes 14 thresholds enabled by default and about 415 telemetry streams across accelerators and cooling. Cooling support covers suppliers including Accelsius, GRC, nVent and ZutaCore.

Park Place positions the update as part of its broader software business serving organisations that need central oversight of complex infrastructure estates. The company said it has annual revenue of USD $1.2 billion, more than 3,000 employees and customers in 180 countries.

Existing Entuity customers with active maintenance contracts can upgrade to the latest version at no additional cost, while new users can deploy directly onto the current release.

"Entuity exists to improve observability and help our customers get the full value out of their infrastructure investment. Version 23.0 P03 takes that mission to the GPU layer, where the stakes and costs are higher. It's giving our customers more insight and control over their environment than ever. AI infrastructure doesn't fail in isolation. GPU health, power delivery, thermal conditions, and cooling performance are all connected. Entuity helps organisations monitor those dependencies from a single operational platform, giving teams earlier visibility into issues that can impact availability, performance, and AI investment returns," said Case Allin, General Manager, Software Solutions, Park Place Technologies.