Case Study: Bing Achieves Real-Time Visual Search with NVIDIA GPUs

Bing deploys NVIDIA technology to speed up object detection and deliver pertinent results in real time

Bing, the search team at Microsoft, wanted to make visual search fast enough to meet user expectations. Their challenge was to detect and match objects in images in real time, which required heavy computation, and they worked with NVIDIA technology, including NVIDIA GPUs and cuDNN, to improve performance and reduce latency.

NVIDIA helped Bing move its object-detection platform from CPUs to Azure NV-series virtual machines with Tesla M60 GPU accelerators, then further optimize it with cuDNN and other model changes. The result was a drop in latency from 2.5 seconds to 40 milliseconds, a 60X performance improvement, more than 80 detectable image categories, and over 90 percent cost savings through smarter filtering and caching.

Bing

Yan Wang

Senior Engineer