Benchmarks
Pixeltable benchmarks
Methods for incremental recompute, computed-column UDF calls, and video frame extraction. Pages publish numbers only from a run that happened, with the machine stated.
Updated
Incremental recompute vs full reindex
Reduced run, 2000 rows, 200 changed. Incremental update 0.059 s wall vs full recompute 0.45 s wall. Stand-in UDF. Intel(R) Xeon(R) Processor, 4 cores, 15.6 GB, Linux 6.12.94+, Python 3.12.3, pixeltable 0.7.21.
Computed-column UDF calls: insert, read, recompute, update
Reduced run, 1000 rows. Insert 1000 UDF calls, stored read 0, full recompute 1000, 100-row update 100. No hosted LLM. Intel(R) Xeon(R) Processor, 4 cores, 15.6 GB, Linux 6.12.94+, Python 3.12.3, pixeltable 0.7.21.
Video frame extraction throughput
Reduced run, 1 clip, 108 frames, 0.24 s wall, 450 frames/s. Intel(R) Xeon(R) Processor, 4 cores, 15.6 GB, Linux 6.12.94+, Python 3.12.3, pixeltable 0.7.21.
The suite lives in benchmarks/. Default scale is reduced. Run python benchmarks/run.py --write. Markdown: /benchmarks.md. Related counts: research.