Looking for an easy way to add AI power? Check out the new AMD Instinct MI350P—a GPU on a PCIe card

The new AMD Instinct MI350P PCIe card lets users deploy and scale generative and agentic AI within existing infrastructure. The card is designed for high performance, affordability, and simple deployment with an open, enterprise-ready AI stack. And it’s being supported by two air-cooled 5U rackmount systems from Supermicro.

Looking for a way to deploy and scale generative and agentic AI on existing server infrastructure? AMD’s got your back with its new AMD Instinct MI350P, a GPU on a PCIe card.

The AMD Instinct MI350P offers a straightforward path to AI. The GPU card simply drops into a standard air-cooled rackmount server.

Who’s it for? Users who need more power for AI inferencing than CPUs can provide, but aren’t ready to invest in dedicated GPU accelerator platforms.

For these users, the new AMD GPU card can be implemented without the need to rebuild AI software stacks, rewrite code, retrain staff, or rearchitect deployments. What’s more, these users can transfer their existing workloads easily, thanks to AMD’s cross-platform interoperability.

Users can also save money. The PCIe card supports the AMD Enterprise AI reference stack, alleviating license fees and restoring local control.

Under the Hood

The new GPU card is based on AMD’s CDNA 4 multichip architecture with 3nm process technology. Each card includes:

AMD has designed this new GPU with open systems. That includes support for Kubernetes GPU Operator for life cycle management. There’s also native support for AI frameworks such as PyTorch.

The AMD GPU also delivers a scalable portfolio with a unified development environment. Namely, the AMD ROCm software ecosystem.

The GPU also offers built-in security features. These include Device Secure Boot, Secure Update and Recovery, and Identity and Attestation.

Supermicro Support

Supermicro has announced that two of its 5U servers are now optimized to deliver the full power of the new Instinct MI350P GPUs. Like the cards themselves, these servers now offer a fast and affordable way to scale GenAI and AI agents without redesigning the data center.

The two Supermicro systems are model numbers AS -5126GS-TNRT2, which supports up to 10 GPUs, and AS -5126GS-TNRT, which supports up to 8.

Otherwise, the two systems have much in common. Both are housed in 5U rackmount enclosures, both are air-cooled, and both are powered by dual AMD EPYC 9004/9005 processors. Key applications for both include AI, deep learning, visualization, multimedia, 3D rendering and HPC.

Do More: