Penguin Solutions Altus XE4318GTSv2-DTC — 4U Direct-to-Chip HGX B300 Server

Specifications

Overview

The Altus XE4318GTSv2-DTC is the direct-to-chip liquid-cooled counterpart to Penguin Solutions' 8U Altus XE8318GTSv2. Moving the same eight NVIDIA HGX B300 GPUs into a 4U chassis halves the rack footprint per node, which is the difference between two and four nodes per 8U of rack space.

Direct-to-chip cooling is what makes the height reduction possible: the B300 baseboard's thermal load is carried into a facility or in-row CDU loop rather than being dissipated into rack airflow. AMD EPYC 9005 or 9004 Turin processors handle host-side processing and PCIe fan-out.

Like its 8U sibling the platform is delivered through Penguin Solutions' OriginAI inference factory programme, with the company holding NVIDIA AI Factory Specialized Partner status since June 2026.

Key Benefits

2x rack density: 4U per eight-GPU node instead of 8U, doubling GPUs per rack against the air-cooled configuration. Lower fan power: removing the GPU thermal load from the airflow path cuts chassis fan energy and acoustic output. Common software path: shares the OriginAI and NVIDIA AI Factory validation of the 8U Altus platform.

Applications

Dense AI inference clusters, reasoning-model serving, LLM training in liquid-cooled halls, colocation deployments that already run a CDU loop, and HPC clusters where rack space rather than power is the binding constraint.

Related Products