Benchmark #114

Download
Another Crab's Treasure BORE/LAVD/LAVD-NO-COMPACTION

This is captured on sched-ext from main (https://github.com/sched-ext/scx/tree/d4a5a629ffc5d01768f6052b1156c9b0479d0f70) with https://github.com/sched-ext/scx/pull/456 and https://github.com/sched-ext/scx/pull/457 backported. This is while in a static scenario while not GPU limited, the system is otherwise idle.

Submitted 11 months ago by summedkibbles36

Specifications
Label OS GPU CPU RAM OS specific
LAVD-NO-COMPACTION Steam Runtime 3 (sniper) NVIDIA GeForce RTX 3070 AMD Ryzen 5 3600 6-Core Processor 17 GB 6.10.1-cb1.0.fc41.x86_64 performance
LAVD Steam Runtime 3 (sniper) NVIDIA GeForce RTX 3070 AMD Ryzen 5 3600 6-Core Processor 17 GB 6.10.1-cb1.0.fc41.x86_64 performance
BORE Steam Runtime 3 (sniper) NVIDIA GeForce RTX 3070 AMD Ryzen 5 3600 6-Core Processor 17 GB 6.10.1-cb1.0.fc41.x86_64 performance

Top runs:

  • Highest FPS: BORE provides the highest mean FPS with 12% more than the next closest run, indicating superior performance in raw frame rate.
  • Smoothest FPS: LAVD-NO-COMPACTION offers the smoothest gameplay experience with the lowest standard deviation and variance in FPS, resulting in the most consistent frame times.
  • Best overall: BORE is the best overall run, balancing higher average FPS (12% more compared to the closest run) with acceptable frame time variability, though slightly higher compared to LAVD-NO-COMPACTION.

Issues:

  • LAVD experiences a 10% lower mean FPS and slightly higher frame time variance compared to BORE, suggesting a less optimal performance.
  • LAVD configuration is similar to the others, making the performance degradation puzzling without further hardware/software inconsistencies.

Summary

This benchmark compared three different configurations (LAVD-NO-COMPACTION, LAVD, BORE) on the same system using sched-ext. Notably, BORE achieved the highest mean FPS, overshadowing both LAVD and LAVD-NO-COMPACTION. However, LAVD-NO-COMPACTION delivered the smoothest gameplay with the least frame time variability. The data suggests that the BORE scheduler implementation provides the most optimal overall performance, merging high frame rates with stable frame times, while LAVD falls behind in both metrics.