computationintensive

Computation-intensive, often written as computation-intensive or compute-intensive, describes tasks or workloads that require substantial computational resources to complete within an acceptable time frame. It is used to distinguish such tasks from those that are primarily memory- or I/O-bound.

Common computation-intensive domains include scientific computing, numerical simulation, cryptography, machine learning model training and inference, data analysis on large datasets, and computer graphics rendering. In CPU-bound tasks, processing speed drives performance; for some workloads, accelerators such as GPUs or TPUs provide a significant speedup due to parallel execution.

Performance assessment uses concepts such as strong scaling (fixing problem size while increasing resources) and weak scaling (increasing both size and resources). Amdahl's law and Gustafson's law frame expectations for speedup and resource utilization. Profiling and benchmarking guide optimization efforts, which may involve algorithmic improvements, vectorization, multi-threading, distributed computing frameworks (MPI, MapReduce), and hardware choices.

Practical considerations include energy consumption, thermal limits, cost, and data movement overhead. Developers aim to maximize computational efficiency while controlling memory access patterns to minimize cache misses and bandwidth bottlenecks. In many environments, compute-intensive workloads are offloaded to specialized hardware or cloud instances to balance performance with cost.

Related topics include high-performance computing, parallel computing, compute-bound workloads, algorithm optimization, and benchmarking.