Hacker News new | ask | show | jobs
by DoctorOetker 115 days ago
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.