Dell, AMD Expand On-Prem AI Platform with Instinct MI350P GPU Support

Why It Matters

The partnership gives enterprises a high‑performance, cost‑effective on‑prem AI option that preserves data control while competing with cloud‑centric and NVIDIA‑driven solutions. It accelerates adoption of generative and agentic AI in regulated industries where latency and security are critical.

Key Takeaways

• Dell PowerEdge XE7745 now supports AMD MI350P GPUs.
• Up to 4,600 TFLOPs peak performance per GPU.
• 144 GB HBM3e memory, highest PCIe capacity today.
• No data‑center redesign needed; drop‑in air‑cooled servers.
• Open stack integrates PyTorch, TensorFlow, vLLM without licensing fees.

Pulse Analysis

Enterprises are moving AI from the cloud to the data center to gain tighter latency control, data sovereignty, and cost predictability. Dell’s latest PowerEdge XE7745 and R7725 servers now ship with AMD’s Instinct MI350P PCIe GPUs, delivering up to 4,600 teraflops of mixed‑precision performance and a record‑setting 144 GB of HBM3e memory. The drop‑in design fits existing air‑cooled racks, eliminating the need for specialized cooling or power upgrades. By pairing AMD’s GPU architecture with Dell’s proven server ecosystem, the solution offers a high‑density, cost‑effective path for scaling generative and agentic AI workloads on‑premises.

Beyond raw horsepower, the Dell AI Platform leverages AMD’s Enterprise AI Suite, ROCm drivers and Inference Server to provide an open, license‑free software stack that integrates seamlessly with PyTorch, TensorFlow, vLLM and other popular frameworks. This reduces the engineering effort required to port models and enables organizations to move from pilot projects to production without extensive code rewrites. The modular architecture lets customers start with a modest GPU count and incrementally increase density, while the same validated firmware and security posture remains intact across upgrades.

The announcement positions Dell and AMD against rivals such as NVIDIA‑based DGX systems and HPE’s GreenLake AI offerings, emphasizing flexibility and total cost of ownership. As foundation models grow in size, the ability to run inference on a PCIe accelerator with 144 GB of HBM3e will become a differentiator for latency‑sensitive applications like real‑time recommendation engines and autonomous agents. Analysts expect the partnership to accelerate on‑prem AI adoption in regulated sectors, where data residency and security concerns keep workloads out of public clouds.