Tensor kernels

Tensor cores are specialized hardware blocks of a GPU. They were first introduced by NVIDIA in the Volta architecture. Due to high performance and process acceleration, tensor cores have become an important part of various forms of data processing and machine learning algorithms.

The mechanics of tensor cores
Kernels perform multiplication and accumulation of mixed precision matrices. Simply put, this means performing multiple calculations at the same time. This capability is an advantage for artificial intelligence (AI), deep learning, and neural network training, where fast and efficient data processing is of paramount importance.

Comparison of tensor and traditional kernels

  • Parallel processing. Unlike traditional kernels that perform one operation at a time, tensor kernels can perform hundreds of operations simultaneously.
  • Calculations with mixed accuracy. Tensor kernels are able to work with calculations of mixed precision. This ability is not available in traditional cores.
  • Optimized for AI. Tensor cores are specifically designed to accelerate AI and deep learning tasks, unlike traditional cores.

Application of tensor cores
Accelerating the learning and processing of complex neural networks by accelerating matrix operations has contributed to advances in various fields of artificial intelligence, including natural language processing, image recognition, and autonomous vehicles.

We use cookies to optimise website functionality and improve our services. To find out more, please read our Privacy Policy.
Cookies settings
Strictly necessary cookies
Analytics cookies