NVIDIA GeForce RTX 2070 OpenCL, CUDA, TensorFlow GPU Compute Benchmarks
What many of you are most interested in... TensorFlow benchmarks with the GeForce RTX 2070. TensorFlow was running within Docker using the NVIDIA GPU Cloud images.
With the ResNet-50 model using FP16 precision, the RTX 2070 was 11% faster than a GeForce GTX 1080 Ti and 86% faster than the previous-generation GeForce GTX 1070.
On a performance-per-Watt basis, the GeForce RTX 2070 is well ahead of the previous-generation Pascal and Maxwell parts.
When running ResNet-50 with FP32, the RTX 2070 performance falls behind the GTX 1080 Ti... Likely due to the tensor cores not coming into play. But still even with FP32, the RTX 2070 is well ahead of the GTX 1080.
The performance-per-Watt at FP32 with the ResNet-50 model was on-par with the GTX 1080 Ti and a 34% improvement over the GTX 1070.