 | This article may rely excessively on sources too closely associated with the subject, potentially preventing the article from being verifiable and neutral. Please help improve it by replacing them with more appropriate citations to reliable, independent, third-party sources. (March 2021) (Learn how and when to remove this message) |
Intel's Deep Learning Boost (DL Boost) is a marketing name for instruction set architecture (ISA) features on the x86-64 designed to improve performance on deep learning tasks such as training and inference.
Features
DL Boost consists of two sets of features:
- AVX-512 VNNI, 4VNNIW, or AVX-VNNI: fast multiply-accumulation mainly for convolutional neural networks.
- AVX-512 BF16: lower-precision bfloat16 floating-point numbers for generally faster computation. Operations provided include conversion to/from float32 and dot product.
DL Boost features were introduced in the Cascade Lake architecture.
A TensorFlow-based benchmark run on the Google Cloud Platform Compute Engine shows improved performance and reduced cost compared to previous CPUs and to GPUs, especially for small batch sizes.
Notes
- "Intel Deep Learning Boost" Product Overview , p. 3
- Samantha Gurriero, "Machine Learning Optimisation: What is the Best Hardware on GCP?", Datatonic,
External links
- Deep Learning Boost at Intel
- Andres Rodrigues et al., "Lower Numerical Precision Deep Learning Inference and Training", Intel White paper
- Intel and ML (2017), from Intel's Developer Relations Division