Everything You Need To Know About The NVIDIA Jetson TX1 Performance

Lastly are benchmarks of Caffe AlexNet, NVIDIA's preferred benchmark for the Jetson TX1 as it fits the use-case of those deploying the Tegra X1 on drones and other portable devices for deep learning purposes. The Phoronix Test Suite's new Caffe test profile provides a CUDA+cuDNN build as well as a CPU-only build for non-NVIDIA deployments. The test profile times 100 iterations of the stock AlexNet model.

As you can see from the results, Caffe AlexNet with CUDA+cuDNN is about twice as fast on the Jetson TX1 as it was on the Jetson TK1. The Tegra X1 thanks to leveraging the Maxwell GPU was much faster than the CPU-only version run on the Intel Xeon E5-2687W v3 Haswell-E system and the Core i5 6600K Skylake desktop.