DL Boost

DL Boost is a marketing term used by Intel to describe a set of technologies designed to accelerate deep learning workloads on their CPUs. The term encompasses specific instruction sets and architectural features implemented in certain Intel processors.

The primary component of DL Boost is the Vector Neural Network Instructions (VNNI), also known as INT8 (8-bit integer) acceleration. VNNI significantly accelerates inference workloads by allowing the processor to perform more operations per clock cycle on 8-bit integer data, which is a common data type used in quantized neural networks. Before VNNI, deep learning inference was often performed using higher-precision floating-point numbers (e.g., FP32), requiring more computational resources and power.

DL Boost, through VNNI, enhances the performance of deep learning inference tasks such as image recognition, natural language processing, and object detection. By utilizing 8-bit integer arithmetic more efficiently, processors with DL Boost can achieve higher throughput and lower latency in these applications. The technology is typically utilized through optimized software libraries and frameworks that can leverage the new instructions.