While AMD's clout in the GPU market pales in comparison to its dominance in the CPU market, the company is brimming with confidence regarding its next - generation AI GPU, the MI450. It boldly asserts that this GPU will deliver all - encompassing leading - edge AI performance.

At the Goldman Sachs Communacopia+ Technology Conference, Forrest Norrod, the Executive Vice President of AMD's Data Center Solutions Business Unit, unveiled the development plans for the MI450. He emphasized that the MI450 will feature a 128 - GPU interconnection design. By leveraging Infinity Fabric over Ethernet, it will achieve a single GPU bidirectional bandwidth that surpasses 1.8TB/s. This is a substantial leap compared to NVIDIA's VR200 NVL144, which only offers a 72 - GPU interconnection solution.

AMD has set its sights on making the MI450 the ultimate solution in the market for a wide range of AI tasks, including training, inference, distributed inference, and reinforcement learning. The company aims to outshine not only NVIDIA's current Blackwell architecture but also its upcoming next - generation Rubin GPU.