Benchmark #65

Download
Automated BG3 benchmark at 2024-08-19 09:31:33 UTC

Automated Baldurs Gate 3 benchmark, on Steam Deck LCD 512GB, at 2024-08-19 09:31:33 UTC. For more information, see https://github.com/erkexzcx/FlightlessSomething-auto/actions/runs/10450353353. LAVD comparison - PR #507 vs PR #508 vs main branch.

Submitted 2 months ago by erikas2

Specifications
Label OS GPU CPU RAM OS specific
lavd-main CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.4-2-cachyos-deckify-lto schedutil
lavd-pr507 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.4-2-cachyos-deckify-lto schedutil
lavd-pr508 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.4-2-cachyos-deckify-lto schedutil
## Top runs: * **Highest FPS**: `lavd-pr507`. It delivers the highest average FPS with a slight edge over the others. * **Smoothest FPS**: `lavd-pr507`. This run has the lowest standard deviation and variance in FPS, making it the most stable. * **Best overall**: `lavd-pr507`. This run not only achieves the highest average FPS but also maintains the smoothest performance, providing the best overall experience. ## Issues: * The run comparing to another code pull request experienced a significant deviation in GPU and CPU temperatures. These were considerably lower, indicating possible underperformance or throttling. * The same run had noticeably higher variance in GPU core clock speeds, showing less stability which could affect overall performance consistency. * Furthermore, the lower GPU VRAM usage suggests inefficiency compared to the other runs. ## Summary This benchmark focuses on comparing different implementations of the `LAVD` component within the same hardware and software environment. The comparison was mainly between the main branch and two pull requests against it. Among the evaluated runs, the second pull request (`lavd-pr507`) showed the best performance by achieving the highest average FPS and demonstrating the smoothest experience with the lowest standard deviation and variance. However, the first pull request had notable fluctuations in GPU temperatures and core clock stability, indicating potential inefficiencies. Overall, the benchmarking highlights the slight but significant performance gains achievable by applying specific updates to the `LAVD` component.