Gigabyte G292-Z44-AAE1 4U Multi-Node GPU Cluster Server EPYC 9654

The Gigabyte G292-Z44-AAE1 is a purpose-built 4U multi-node GPU cluster server engineered around the AMD EPYC 9654 (Genoa) processor, delivering 96 cores and 192 threads of x86 compute per node with support for DDR5 memory across 12 DIMM slots. The platform is architected to host up to 10 PCIe 5.0 x16 GPU accelerators per chassis, enabling extreme GPU density critical for large-scale deep learning training, inference serving, and HPC simulation workloads. The PCIe 5.0 interconnect fabric doubles the available bandwidth versus previous-generation deployments, ensuring that GPU-to-GPU and GPU-to-host data paths are not bandwidth-constrained even with the highest-throughput accelerators available in 2025.

The G292-Z44-AAE1 integrates 400GbE fabric readiness, allowing data center operators to connect each node into high-bandwidth spine-leaf network topologies using QSFP-DD or OSFP form-factor NICs without chassis modifications. The server's multi-node architecture allows independent node management and workload isolation within a shared physical footprint, maximizing rack density while preserving operational flexibility. Dual OCP 3.0 mezzanine slots, multiple NVMe/SATA storage bays, and a dedicated IPMI/BMC management port round out a platform designed for 24/7 mission-critical cluster operation.

Targeting AI infrastructure teams deploying GPU clusters at scale—whether for large language model (LLM) training, recommendation engines, autonomous vehicle simulation, or scientific computing—the G292-Z44-AAE1 is validated with leading GPU accelerators including NVIDIA H100/H200 and AMD Instinct MI300X class cards. Gigabyte's MegaRAC SP-X BMC provides out-of-band management, BIOS configuration, sensor telemetry, and KVM-over-IP, enabling fully remote lifecycle management suited to hyperscale and colocation environments launching in Q4 2025.

Ideal for

• Large language model (LLM) pre-training and fine-tuning requiring maximum GPU density and PCIe 5.0 bandwidth within a compact 4U chassis
• Distributed AI inference clusters serving real-time requests with multiple GPU accelerators per node and low-latency 400GbE fabric interconnects
• High-performance computing (HPC) simulation workloads in scientific research, genomics, and computational fluid dynamics leveraging EPYC 9654's 96-core throughput
• Multi-tenant GPU cloud infrastructure where independent multi-node management enables isolated workload partitioning within shared physical hardware
• Autonomous vehicle and robotics simulation pipelines that require sustained GPU compute alongside high-core-count CPU preprocessing
• Enterprise recommendation system training and real-time serving combining dense GPU acceleration with DDR5 memory bandwidth for embedding table operations

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.