Compute Engines: Cross-Category Organization and Trade-offs

4.10.1The Balance Among Scalar/Vector/SIMD/SIMT/Matrix Engines#

4.10.2Systolic Array, Tensor Core, Vector MAC, and Configurable Compute Units#

4.10.3Matrix Throughput, Non-GEMM Operators, and Control-Flow Coverage#

4.10.4Embedding, Gather/Scatter, and Dedicated Acceleration Units#

4.10.5Static Scheduling, Dynamic Scheduling, and Data-Triggered Execution#

4.10.6Arithmetic Intensity, Compute Utilization, and Area/Energy Efficiency#

4.10.7Programming Flexibility, Compiler Burden, and Degree of Specialization#