Ambient Occlusion Benchmark
For our paper Visionaray: A Cross-Platform Ray Tracing Template Library, we reported results gathered with our simple AO benchmark program implemented with Visionaray. Going to update the results from time to time here for new hardware platforms.
Platform | Conference Room | Fairy Forest | Sibenik Cathedral |
NVIDIA RTX A6000 (RTX=OFF) | 588.3 (586.2) | 291.9 (294.2) | 597.3 (588.6) |
NVIDIA RTX 3060Ti (RTX=OFF) | 558.0 (580.8) | 190.4 (190.5) | 286.7 (578.4) |
NVIDIA GTX 1080Ti | 290.7 (290.9) | 143.4 (143.5) | 288.5 (290.7) |
Intel KNL with 64x4 threads | 254.0 (273.2) | 130.0 (130.4) | 247.7 (268.2) |
AMD Ryzen 7 4700G, 8x2 threads | 195.7 (214.0) | 92.4 (93.7) | 168.6 (187.9) |
NVIDIA GTX 980Ti | 172.9 (228.3) | 73.4 (83.3) | 139.1 (188.5) |
NVIDIA Quadro K6000 | 142.4 (189.1) | 63.1 (71.1) | 113.2 (142.9) |
NVIDIA Titan (1st gen, Kepler) | 114.4 (143.3) | 55.5 (56.9) | 81.5 (114.4) |
16-core dual socket Xeon E5-2690 | 102.0 (117.0) | 55.0 (57.1) | 94.2 (105.6) |
Apple MBP M2 Max, 2023, 8 threads (perf.) + 4 threads (eff.) | 96.3 (108.2) | 48.2 (50.8) | 81.9 (91.9) |
NVIDIA GTX 970M | 83.6 (107.1) | 32.7 (36.4) | 63.9 (87.9) |
Intel Core i7 6800K, 6x2 threads | 80.1 (89.4) | 40.6 (41.7) | 68.6 (77.8) |
Apple MBP M1, 2020, 4 threads (perf.) + 4 threads (eff.) | 47.7 (53.4) | 24.5 (26.1) | 40.8 (45.8) |
Apple MBP Mid 2014 2.8 GHz Core i7, 4x2 threads | 41.8 (46.2) | 20.8 (21.0) | 34.7 (39.4) |
Nehalem Core i7 920, 2.67 GHz, 4x2 threads | 20.7 (21.9) | 11.4 (11.9) | 15.9 (18.5) |
Raspberry PI3 with ARM NEON | 2.0 (2.3) | 1.2 (1.3) | 1.7 (1.9) |
Results are in Mrays/s, 1024 x 1024 px. images, averaged over 250 runs, 8 AO rays. We report results for the binned BVH builder, and the SBVH builder in parentheses.
A tarball containing the benchmark script can be downloaded from here. Results for hardware platforms not listed here are highly welcome. Please send those to me (Stefan Zellmann, info (at) szellmann (dot) de), along with the output of cat /proc/cpuinfo
or sth. similar, and a brief note which benchmark you ran (CUDA, SSE, AVX).
Contributors
Thanks go to: - Martin Aumüller for the i7 6800K results
- Matthias Hellmann for providing the GTX 980Ti and GTX 970M results