Penguin Solutions Altus XE8318GTSv2 — 8U AMD EPYC Turin HGX B300 Server

Specifications

Overview

The Altus XE8318GTSv2 is Penguin Solutions' 8U eight-GPU platform pairing AMD EPYC 9005 or 9004 series Turin processors with an NVIDIA HGX B300 baseboard. It is aimed squarely at reasoning and inference workloads where the B300's memory capacity per GPU is the limiting factor rather than raw matrix throughput.

Four PCIe Gen5 x16 FHHL double-width slots remain available for networking, storage or DPU cards on top of the eight SXM GPUs, so the node can carry both a high-radix east-west fabric and dedicated north-south offload without a separate riser shelf.

Penguin Solutions builds and validates the node as part of its OriginAI inference factory platform, and became an NVIDIA AI Factory Specialized Partner in June 2026 - which means the platform is delivered with cluster-level orchestration, not just bare metal.

Key Benefits

B300 memory per GPU: 8-GPU HGX B300 density in an 8U envelope suits long-context reasoning where KV working sets dominate. Turin compute headroom: EPYC 9005/9004 class CPU cores keep host-side data movement and preprocessing from bottlenecking the GPUs. Four spare Gen5 x16 slots: networking and DPU add-ons can be integrated without sacrificing a GPU position.

Applications

Large language model reasoning and inference, model fine-tuning on 8-GPU nodes, retrieval-augmented generation pipelines, scientific simulation with GPU-accelerated solvers, and enterprise AI factory deployments managed through OriginAI.

Related Products