NVIDIA / NVIDIA/cuvs

[QST] CAGRA build time is nearly identical across float, half, and int8 — is this expected?

Open
#1,857 10 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

question
Dominant language
Cuda
Stars
854
Forks
236
Avg merge
3d 3h
Merged PRs (30d)
62

Description

What is your question?
I'm benchmarking CAGRA build performance with different data types (float, half, int8) on the same dataset. I expected int8 to be significantly faster than float since it uses 4x less memory per element, but the build times are nearly identical across all types (~5-7% difference).

Is this expected behavior, or is there a performance issue?

Steps/Code to reproduce bug
I wrote a simple benchmark that loads the Cohere-1M dataset (1M vectors, 768 dims), converts it to different types, and calls cuvs::neighbors::cagra::build() for each.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the reported Cohere-1M benchmark and its calls to cuvs::neighbors::cagra::build() for float, half, and int8. Reproduce the timing comparison, then determine whether the similar build times are expected or indicate a reproducible performance defect; done means documenting the finding or defining the defect clearly.

Written by the indexing model from the issue text.

Assessment

Domain
performance, search
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.