The random access memory models is not really representative of ML workloads (both training and inference), where multiplying large tensors result in predictabl

The random access memory models is not really representative of ML workloads (both training and inference), where multiplying large tensors result in predictable memory access patterns.