Skip to main content

Benchmarks

This page has two kinds of evidence, and they are not comparable with each other:

  • Algeon relational benchmarks below are Algeon's own TPC-H and TPC-DS runs. CPU and GPU rows come from the same bench harness, the same release build, and the same fixture data generated by fixture/fixture.sh. Each section states its host, run configuration, and the artifact it was generated from.
  • Graph algorithm benchmarks are published upstream by the RAPIDS cuGraph project for its own Python and NetworkX surfaces. Algeon does not re-run them, and they say nothing about the cugraph_* SQL functions' timing in Algeon.

All numbers are single runs on one machine at one revision. Treat them as evidence about that configuration, not as a guarantee for another dataset, GPU, or query mix.

Algeon TPC-H SF100​

Generated from target/bench/tpch_sf100_gpuonly_localkvikio_compat_on_nocache_warm1_iter1_20260720T202206Z/native_tpch_benchmark.tsv on 2026-07-20. The per-query rows and chart are rendered from website/src/data/tpchSf100Benchmark.ts, which is regenerated from that TSV by website/scripts/generate_tpch_sf100_benchmark_ts.py.

  • Environment: AWS EC2 g7e.4xlarge, 16 vCPUs, 128 GiB system memory, local NVMe fixture storage, RTX PRO 6000.
  • Run: SF100,1 warmup / 1 measured iteration, source cache disabled, LocalKvikio parquet reads.
  • Result: CPU 77.4s, GPU 26.2s, CPU / GPU 3.0x, peak reservation 54.6 GiB.
  • KvikIO: local cuDF/libcudf file reads with cache disabled; this run forced KvikIO compatibility I/O with KVIKIO_COMPAT_MODE=ON. On this g7e.4xlarge host, OFF enters the cuFile path, but cuFile compatibility mode can still fall back; it does not prove native GDS.

Wall time, measured iteration mean.

Algeon TPC-DS SF100​

Generated from target/bench/tpcds_sf100_gpuonly_localkvikio_compat_on_nocache_warm1_iter1_20260720T225024Z/ (cpu_tpcds_three_way_bench.tsv and gpu_tpcds_three_way_bench.tsv) on 2026-07-20. The rows and chart are rendered from website/src/data/tpcdsSf100Benchmark.ts, regenerated by website/scripts/generate_tpcds_sf100_benchmark_ts.py.

  • Environment: AWS EC2 g7e.4xlarge, 16 vCPUs, 128 GiB system memory, local NVMe fixture storage, RTX PRO 6000.
  • Run: SF100 full suite (99 queries), 1 warmup / 1 measured iteration, profile sf100_gpu_throughput, LocalKvikio parquet reads.
  • Result: CPU 379.4s, GPU 101.7s across 99/99 successful native queries, CPU / GPU 3.7x, peak reservation 43.0 GiB. Row counts matched on 99/99 queries.
  • KvikIO: local cuDF/libcudf file reads with KVIKIO_COMPAT_MODE=ON. On this g7e.4xlarge host, OFF still may enter cuFile compatibility mode without native GDS.

Wall time for the measured iteration. GPU bars use successful native time; failed attempts would still show wall time in gray. Y-axis skips 20–120s so short queries keep readable CPU vs GPU heights; Q72 and other long outliers still render above the break.

Scroll horizontally to inspect all 99 queries.

Upstream cuGraph benchmarks​

The RAPIDS project publishes official cuGraph benchmarks for its Python and nx-cugraph APIs. They use a different harness, dataset, and hardware from the Algeon runs above.