AMD Instinct MI325X 288GB OAM for Dell PowerEdge XE9680

The AMD Instinct MI325X 288GB OAM for Dell PowerEdge XE9680 is a next-generation accelerator built on AMD's CDNA 3 architecture, delivering exceptional compute density and memory bandwidth for the most demanding AI and HPC workloads. Featuring 288GB of HBM3E memory with an aggregate bandwidth of 6.0 TB/s, the MI325X represents a significant step forward in high-capacity accelerator design, enabling larger models and datasets to reside entirely on-device without memory bottlenecks. The OAM (OCP Accelerator Module) form factor integrates natively into the Dell PowerEdge XE9680 platform, which supports up to eight OAM accelerators in a single 8U chassis with high-bandwidth NVLink-class interconnects replaced by AMD Infinity Fabric links for peer-to-peer communication.

The MI325X is engineered around AMD's CDNA 3 compute units, providing peak FP64 matrix performance, FP32, FP16, and FP8 data type support — the latter being essential for large language model (LLM) inference and training at scale. The accelerator supports AMD ROCm open software ecosystem, offering compatibility with HIP, OpenCL, and frameworks such as PyTorch and TensorFlow. Its OCP-compliant OAM form factor and integration within the PowerEdge XE9680 allow for direct liquid-cooling readiness and enterprise-class thermal management, ensuring sustained boost performance in dense rack deployments across 24/7 datacenter operations.

Designed specifically for enterprise AI infrastructure, the AMD Instinct MI325X 288GB OAM paired with the Dell PowerEdge XE9680 addresses the requirements of large-scale generative AI training, scientific simulation, and real-time inference serving. Dell's validated platform engineering ensures firmware compatibility, telemetry integration through iDRAC 9, and full support for OpenManage tooling, making this accelerator suitable for regulated industries and mission-critical AI pipelines in UAE, GCC, EMEA, and APAC deployments.

Ideal for

• Large language model (LLM) pre-training and fine-tuning requiring high HBM3E memory capacity to load multi-billion parameter models on a single node
• Generative AI inference serving at scale, leveraging FP8 precision support for high-throughput, low-latency token generation in production environments
• High-performance computing (HPC) simulations in computational fluid dynamics, molecular dynamics, and climate modelling using FP64 double-precision compute
• AI-driven medical imaging and genomics workloads demanding both large memory footprints and sustained compute throughput in regulated datacenter environments
• Multi-tenant AI platform deployments in enterprise and cloud-adjacent datacenters where OAM form factor density and ROCm software compatibility reduce operational complexity
• Data analytics and deep learning recommendation model (DLRM) training where 288GB on-device HBM3E eliminates frequent host-to-device memory transfers

Technical specifications

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