NewsBite analysis · AI, Hardware
wccftech
Packs CDNA 4 GPU With 4.6 PFLOPs AI Compute, 144 GB HBM3E at 600W
AMD has introduced the MI350P, its first PCIe GPU designed for AI applications in four years. With its CDNA 4 architecture, the MI350P provides up to 4.6 PFLOPs of AI compute power and 144 GB of HBM3E memory, offering a competitive alternative in the enterprise market. Designed for easy deployment in existing data infrastructures, it supports various AI precision formats, making it appealing to organizations looking to scale their AI capabilities without significant infrastructure investment. The launch underscores AMD's commitment to expanding its AI compute portfolio amid increasing demand for powerful AI solutions.
Act nowConfidence: HighImmediate horizon
Published 7 May 2026, 16:21 UTC3 min read
Visit original article ↗Lead image — wccftech
Summary
TL;DR
- 4.6 PFLOPs peak AI compute performance
- 144 GB HBM3E memory, 4 TB/s memory bandwidth
- First PCIe GPU from AMD in four years
- Designed for easy integration into existing data centers
- Supports various AI precision formats
Interesting facts
- The MI350P is built using TSMC's 3nm and 6nm processes, featuring 73 billion transistors.
- It offers a dual-slot design suitable for standard air-cooled servers, emphasizing ease of use.
- AMD positions the MI350P against NVIDIA's H200 NVL, highlighting intense competition in the AI GPU market.