Dell PowerEdge Servers Adopt AMD Instinct MI350P PCIe GPUs for On-Prem AI Scaling

Dell Technologies and AMD have announced an expansion of their enterprise hardware collaboration, introducing high-density artificial intelligence processing capabilities directly into standard, air-cooled data centre footprints. Starting in July 2026, Dell PowerEdge XE7745 and R7725 servers will provide native validation and hardware support for the newly unveiled AMD Instinct MI350P PCIe GPUs.

The hardware launch is accompanied by an expansion of the Dell AI Platform with AMD, a comprehensive infrastructure ecosystem designed to help organisations transition machine learning workloads out of isolated pilot sandboxes into production-grade enterprise deployment.

Solving the Data Centre Retrofitting Bottleneck

A significant operational barrier to enterprise AI scaling has been the infrastructural demand of modern graphics accelerators. Many current-generation AI chips are designed as OAM socketed modules on universal baseboards, which pull near-kilowatt power levels and necessitate custom server chassis, high-amperage power delivery systems, and liquid-cooling loops.

The integration of the AMD Instinct MI350P PCIe GPU addresses this limitation by adapting top-tier silicon performance into a standard, full-height, dual-slot PCIe card form factor. This architecture enables corporate IT teams to scale processing densities within the limits of their existing standard air-cooled racks, completely bypassing the capital expenditures and operational downtime associated with facilities redesign.

Technical Performance Metrics and Core Architecture

The PowerEdge servers, augmented by the MI350P PCIe architecture, deliver high-throughput compute designed primarily for distributed inference, Retrieval-Augmented Generation (RAG) data pipelines, and multi-layered agentic workflows:

• Compute Performance: Delivers up to 4,600 peak teraflops of processing capacity utilising the advanced MXFP4 data type standard, striking an efficient balance between execution speed and mathematical precision.

• Memory Architecture: Features 144GB of high-bandwidth HBM3e memory per individual card. This matches the highest current capacity configuration available within a PCIe accelerator form factor, providing the operational headroom required to host large, token-heavy language models on-premises.

• Chassis Configuration: The 4U air-cooled PowerEdge XE7745 supports up to 8 double-wide or 16 single-wide PCIe accelerators alongside up to 192 AMD EPYC CPU cores, allowing enterprise architects to adjust core-to-GPU ratios to align with specific data centre workloads.

Open Software Abstraction Layer

To eliminate restrictive developer ecosystems and vendor lock-in, the modular infrastructure platform leverages a completely open-source application stack.

The architecture utilises the AMD Enterprise AI Suite, AMD ROCm, and the AMD Inference Server out of the box, integrating seamlessly with mainstream machine learning frameworks such as PyTorch, TensorFlow, and vLLM. This software approach allows software developers to migrate existing cloud-built models onto local bare-metal configurations with minimal source code refactoring and zero ongoing licensing fees.

For ultra-demanding data centre configurations, Dell also highlights its high-tier PowerEdge XE9785 arrays. Purpose-built for core foundation model development and large-scale industrial inference, these top-end systems support high-wattage socketed AMD Instinct MI355X GPUs and 5th Gen AMD EPYC processors.

Ecosystem Perspectives

Venkat Sitaram, Senior Director and Country Head of the Infrastructure Solutions Group at Dell Technologies India, emphasised the business need for structural simplicity as enterprises repatriate workloads from public clouds:

"As organizations move from AI experimentation to enterprise-scale deployment, they need infrastructure that combines performance, scalability and simplicity. Dell Technologies and AMD are helping customers accelerate this journey by enabling generative and agentic AI workloads to run efficiently within existing data center environments, without compromising on control, security or flexibility. Together, we are helping enterprises operationalise AI faster and unlock greater business value from their AI investments."

Read More:

How AMD's AI strategy is opening new growth avenues for channel partners

Partner Pulse: Shivaami Cloud Services| Cloud Partner (India)

How Unicorn Infosolutions is accelerating Apple's retail growth in India

Why India's IT Channel Is Moving from Cloud Resale to Managed Services