Overview

XVector is a high-performance vector search library for XCENA’s Computational Memory (MX1). It offloads compute-intensive approximate nearest neighbor (ANN) search to the device, delivering high throughput for large-scale vector workloads.

Supported Algorithms

Algorithm Description Use Case
IVF-Flat Inverted file index with exact distance computation Best accuracy within IVF family
IVF-RaBitQ Inverted file index with randomized binary quantization Ultra-compact, fast approximate search
KNN Exact (Flat) Brute-force exact nearest neighbor search 100% recall ground-truth and small-to-medium datasets

Key Features

  • Device-accelerated search – Runs ANN search on MX1 via PXL runtime
  • Faiss drop-in replacement – Swap to XFaiss to accelerate search on MX1 with existing .faiss index files

See Getting Started to install XVector and run your first search.