Skip to content
MRHI

Performance

These charts were generated from a self-run benchmark using a third-party, open-source benchmark suite based on the ann-benchmarks repository.

They are included here as benchmark artifacts for the public docs. They are not interactive, and they reflect the benchmark runs that were generated for this site.

QPS

Section titled “QPS”

10k vectors

Section titled “10k vectors”

QPS benchmark at 10k vectors

100k vectors

Section titled “100k vectors”

QPS benchmark at 100k vectors

Full run

Section titled “Full run”

QPS benchmark across the full run

P99 Latency

Section titled “P99 Latency”

10k vectors

Section titled “10k vectors”

P99 latency benchmark at 10k vectors

100k vectors

Section titled “100k vectors”

P99 latency benchmark at 100k vectors

Full run

Section titled “Full run”

P99 latency benchmark across the full run

Build Speed

Section titled “Build Speed”

10k vectors

Section titled “10k vectors”

Build speed benchmark at 10k vectors

100k vectors

Section titled “100k vectors”

Build speed benchmark at 100k vectors

Full run

Section titled “Full run”

Build speed benchmark across the full run

Notes

Section titled “Notes”
  • These charts come from a self-run benchmark workflow
  • The benchmark suite used was third-party and open-source
  • The benchmark basis was the ann-benchmarks repository
  • MRHI itself is compatible with CPUs, CUDA GPUs, and Apple Metal