NVIDIA

A vendor's benchmark chart plots QPS against recall@10. What should you look for, and what claim should make you skeptical?

Spot the conceptual error: 'We're using PQ instead of HNSW for our vector index because PQ is faster.'

Walk through how Product Quantization compresses a 1024-dim float32 vector down to 32 bytes, step by step.

PQ-based distance computation outruns full float32 distance even after decoding the vectors. Why?

Predict the shape of recall and latency on an IVF index of 10M vectors as nprobe sweeps from 1 to 50.

Spot the error in this candidate's description of IVF's two main parameters.

Estimate the RAM cost of an HNSW index over 100M vectors at 1024 dimensions with M=32, including graph overhead.

As M grows in an HNSW index, four things move at once. Identify what climbs versus what slows.

Explain what NVIDIA CAGRA buys you over CPU HNSW and the operational catch that decides whether you should reach for it.

Same block, two regimes: training parallelism versus inference autoregression