The growing demand for AI – particularly the computation-intensive training workloads associated with generative (gen) AI models – is driving the adoption of specialized servers with processors known as accelerators. GPUs are the most common type of accelerator for gen AI, with their parallel processing capabilities, versatility across applications, high memory bandwidth, and scalability. GPUs are delivering performance gains with every new generation, but power consumption grows as well (specified as thermal design power or TDP).