NVIDIA GeForce RTX 2080 Ti Shows Very Strong Compute Performance Potential

First up are a few of the CUDA benchmarks on the tested NVIDIA cards. (ROCm's HIP wasn't attempted for this comparison for CUDA codes running on Radeon GPUs due to various obstacles encountered in trying to deliver this timely GPU compute comparison.)

In the degridding test with ASKAP, the RTX 2080 Ti is 64% faster than the GTX 1080 Ti (or 40% for gridding). This is a larger generational improvement going from Pascal to Turing than from Maxwell to Pascal.

With a basic CUDA mini nbody test even, the GeForce RTX 2080 Ti wound up running about twice as fast as the GTX 1080 Ti!

The SHOC OpenCL benchmark achieved 16.5 TFLOPS of single-precision compute power in this particular test, a 25% improvement over the GTX 1080 Ti. In this test and a few other OpenCL tests, the ROCm 1.9 stack had problems running on the RX Vega 64 graphics card.

During the SP FLOPS test, the GeForce RTX 2080 Ti was pulling about 20 Watts on average more than the GTX 1080 Ti.

But on a performance-per-Watt basis, the RTX 2080 Ti still came out ahead by 14%.