Benchmark #45

Download
Automated BG3 benchmark at 2024-08-07 09:45:17 UTC

Automated Baldurs Gate 3 benchmark, on Steam Deck LCD 512GB, at 2024-08-07 09:45:17 UTC. For more information, see https://github.com/erkexzcx/FlightlessSomething-auto/actions/runs/10280938519. bpdfland comparison of --slice-us values

Submitted 3 months ago by erikas2

Specifications
Label OS GPU CPU RAM OS specific
bpfland-s1000 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
bpfland-s16000 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
bpfland-s20000 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
bpfland-s2000 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
bpfland-s4000 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
bpfland-s500 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
bpfland-s8000 CachyOS AMD Custom GPU 0405 (RADV VANGOGH) AMD Custom APU 0405 16 GB 6.10.3-2-cachyos-deckify-lto schedutil
## Top runs - **Highest FPS**: `bpfland-s500` with a mean FPS approximately 0.3% higher than the other runs. - **Smoothest FPS**: `bpfland-s1000` with the lowest FPS standard deviation and variance—about 1% lower than the next closest run. - **Best overall**: `bpfland-s1000` with the best balance of high FPS and smooth performance (lowest FPS variance). ## Issues - Significant FPS issue noted in the `bpfland-s500` run: - Lowest FPS is about 17% lower than other runs. - Although its mean FPS is slightly higher, the significantly high Lowest FPS indicates instability. - This instability is supported by the high variance of GPU core clock (about 3 times higher compared to others) and GPU VRAM used variance (about 1000 times higher). ## Summary All runs employed the same hardware and software settings, with identical Linux kernel, scheduler, GPU, and CPU configurations. Most of the runs deliver fairly consistent performance, with slight variations in FPS and frame time metrics. The `bpfland-s1000` run emerges as the top performer in terms of stability and balanced performance, whereas `bpfland-s500` has an unstable FPS despite its marginally higher mean FPS. This instability might be due to extreme variance in GPU core clock and VRAM utilization, indicating a potential hardware or configuration anomaly during that specific test. The other runs performed similarly to each other with minor differences in FPS and stability metrics within an acceptable range.