AMD releases Instinct MI350P PCIe cards for enterprise AI
Updated
Updated · amd.com · May 7
AMD releases Instinct MI350P PCIe cards for enterprise AI
8 articles · Updated · amd.com · May 7
The dual-slot, air-cooled cards are designed for standard servers, with up to 4,600 peak TFLOPS and 144GB of HBM3E memory.
AMD says they let companies run inference and retrieval-augmented generation workloads on premises without major power, cooling or rack redesigns, and support up to eight cards per system.
The launch targets enterprises seeking alternatives to cloud AI costs and privacy concerns, while extending AMD's open software ecosystem with Kubernetes, PyTorch and inference microservices support.
Can AMD’s new 'drop-in' AI card truly compete without a custom high-speed server design?
Is AMD's open software powerful enough to finally break NVIDIA’s decade-long stranglehold on AI?
AMD promises cheaper on-premise AI, but are the hidden integration costs higher than the cloud?