• Speed vs. accuracy index parameters
• Pick the index family
• Pick the index parameters
• K-means clustering
• Query batches
• Train_shared training mode for PQ training
• System configuration
• Huge memory pages
• NUMA nodes
• Intel MKL library
• Sizes of various internal buffers
• AVX-512 support
• Transposed centroid table for Product Quantizer
• Vector codecs
• Implementation-level
• C++ templates
• CUDA templates
• Feedback