NVIDIA and Oracle have formalized the development of GPU-accelerated vector indexing technology.
This innovation is based on the integration of NVIDIA GPUs and the cuVS open source library with Oracle’s Private AI Service Container and AI Database solutions.
The objective is to transfer to the GPU tasks previously carried out by CPUs, in particular the creation of indexes and vector search. The result is a significant acceleration of processing and better exploitation of massive volumes of data.
The growing challenge of unstructured data
Faced with the explosion of multimodal data — texts, images, videos — companies are finding it difficult to quickly exploit this information.
The creation of vector indexes, essential for AI systems, remains a complex and time-consuming process.
The technology developed by NVIDIA and Oracle aims precisely to reduce this bottleneck, by drastically shortening processing times and improving the overall efficiency of artificial intelligence models.
Already concrete use cases in health
Several industrial applications already illustrate the potential of this innovation. The company Sophia, specializing in medical AI, operates a database of more than 500 million vectors, representing around 3 terabytes.
With GPU acceleration, the time required to create indexes, previously days, has been significantly reduced.
For its part, BioPie relies on Oracle’s cloud infrastructure equipped with NVIDIA GPUs to analyze infectious diseases and propose treatments. By combining database and vector search, the company managed to reduce latency while optimizing costs.
A key building block for large-scale AI
Beyond the medical sector, NVIDIA emphasizes the structuring role of GPU-accelerated vector search.
This technology is establishing itself as a pillar of modern AI infrastructures, particularly for real-time processing and large-scale training and inference environments.
By facilitating the rapid exploitation of unstructured data, it opens the way to new uses in many industrial sectors.
