Found 1 document(s)
Page 1 of 1
-
AI training and inference workloads make GPU power swing repeatedly between compute and communication phases. As a result, the cascaded converters in a datacenter rack keep moving between operating points with different ...