Seibert, “Numba: A llvm-based python jit compiler,” in Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC, 2015, pp.
Kandrot, CUDA by example: an introduction to general-purpose GPU programming. Rocklin, “Dask: Parallel computation with blocked algorithms and task scheduling,” in Proceedings of the 14th python in science conference, 2015, vol. Zaharia et al., “Apache spark: a unified engine for big data processing,” Commun. Schaller, “Moore’s law: past, present and future,” IEEE Spectr., vol. Traditionally applications were designed to run sequentially and therefore dramatically slowing down the execution of the running process. These processors facilitate rapid performance and enable giga floating-point operations per second (GFLOPS). To combat some of these limitations’ CPU vendors have switched to the development and manufacture of multi-core CPUs that enable multithreading tasks. The performance of CPUs has seen limited improvement in recent years which is in stark contrast to the early decades which saw CPU performance double every year (Moores Law).
Architectures based on single CPU’s which are found in all types of computer hardware (Intel/AMD) have started to stall in terms of their performance due to restrictions on the manufacturing process and heat dissipation issues. Although this is a widespread practice for DL applications, historically the training of traditional machine learning models such as SVM’s and RF’s have been restricted to CPU compute. This is achieved by providing users with the ability to execute end-to-end data science pipelines on GPU’s or large-scale CPU based clusters. Accelerated machine learning is an exciting new paradigm in the field of AI which aims to improve the efficiency of training machine learning models.
0 kommentar(er)
