Hello.
I am noticing that the Orin AGX is performing worse in some situations than the Xavier.
Specifically, I ran Phoronix Test Suite on both an Orin AGX devkit (64GB model) and a Xavier devkit on a variety of different power states on both.
Happy to provide the full data from my benchmarks but here’s the key takeaways:
- pts/cuda-mini-nbody: Orin performs better (by a lot - 2x to 3x) at MAX-N. At 30W and 15W Xavier performs better.
- pts/scimark2 (FFT): At MAX-N, Orin performs best. Orin performs better at 30W than at 50W. Xavier performs much better than Orin (3x-4x as fast) at 15W but worse at 30W.
- pts/cachebench: At MAX-N, Orin performs marginally better. At 15W, Xavier performs better than Orin.
- pts/ramspeed: Xavier performs better at all power levels.
- system/openssl: Orin performs about 1.5x better than Xavier at MAX-N and 30W. At 15W, Xavier performs better than Orin.
I also ran a pretty simple benchmark of the two where it computed the max value of a very large array. The Xavier consistently beat the Orin at all power levels.
Is there any way to explain this discrepancy? Can someone from NVIDIA explain how they were able to get the 1.9x figure cited in their description? Am I just doing something wrong here?
