AMD approached to make world's fastest AI supercomputer powered by 1.2 million GPU

AMD was asked to create an AI supercomputer with 1.2 million GPUs, potentially 30x more powerful than Frontier, illustrating growing AI demands.

: AMD has been approached to build an AI supercomputer featuring 1.2 million GPUs, indicating a potential major shift in the AI market. This supercomputer could be 30 times more powerful than the current fastest AI supercomputer, Frontier. The move highlights AMD's growing relevance in a market dominated by Nvidia.

AMD has received inquiries to create an AI supercomputer powered by 1.2 million GPUs, potentially making it the fastest AI supercomputer in the world. This development marks a significant opportunity for AMD, which supplied less than 2% of data center GPUs in 2023.

GM of Datacenter Solutions, Forrest Norrod, disclosed that these inquiries came from serious clients, contemplating spending up to $100 billion on AI training clusters. Current AI training clusters use a few thousand GPUs, whereas this proposed cluster would be 30 times more powerful than Frontier, the largest AI training supercomputer with 37,888 GPUs.

Building such a supercomputer comes with challenges including low latency requirements, high power consumption, and hardware reliability. Despite these hurdles, the push towards advanced AI and machine learning capabilities underscores the significant changes happening in data centers, moving towards handling large, synchronous AI training tasks.

AI supercomputerGPUsData centers