NVIDIA GeForce RTX 2070 OpenCL, CUDA, TensorFlow GPU Compute Benchmarks
As I have only had this graphics card since this morning, this initial article is a quite short and straight-forward look at the GPU compute performance considering most of the other (Windows) reviews on the RTX 2070 over the past two days have just been focused on the gaming performance... But, yes, I will have Linux gaming benchmarks with this EVGA GeForce RTX 2070 graphics card in the next few days -- all of those benchmarks will also be done with the Core i9 9900K.
A variety of OpenCL and CUDA benchmarks were run, including TensorFlow with various models using the NVIDIA GPU Cloud Docker images. During the benchmarking process via the Phoronix Test Suite, it was also monitoring the GPU core temperature as well as the AC system power consumption in real-time for also being able to provide accurate performance-per-Watt metrics. Interfacing with PTS for the AC power consumption readings was a WattsUp Pro via USB interface.
For these GPU compute focused tests, the AMD Threadripper 2990WX system was being used. The graphics cards I tested for this comparison included the:
- Radeon R9 Fury
- Radeon RX 580
- Radeon RX Vega 56
- Radeon RX Vega 64
- GeForce GTX 980
- GeForce GTX TITAN X
- GeForce GTX 1070
- GeForce GTX 1070 Ti
- GeForce GTX 1080
- GeForce GTX 1080 Ti
- GeForce RTX 2070
- GeForce RTX 2080 Ti
Unfortunately I have no Titan V nor any GeForce RTX 2080 so that's why there are no tests with those GPUs. The Radeon tests were done using the latest ROCm 1.9.1 release and due to software compatibility obviously not all of the tests could be run on the Radeon GPUs -- thus this testing is mostly NVIDIA focused in this article, at least until the ROCm ecosystem has matured further and AMD back in the races with a high-end performance GPU to take on the RTX series.
On the NVIDIA side, all of these graphics cards were tested this week using the brand new NVIDIA 410.66 stable Linux driver update and with CUDA 10.0 installed.
Ubuntu 18.04 LTS was running on the system with the Linux 4.18 stable kernel.
Performance-per-dollar graphs are also included based upon the current pricing of the tested GPUs via NewEgg.
With all the basics covered, let's get straight to seeing the GPU compute performance potential of the RTX 2070. Though as a quick reminder if you do enjoy my daily Linux hardware benchmarking, consider showing your support by joining Phoronix Premium or making a PayPal tip. Phoronix Premium provides access to the site ad-free, multi-page articles are shown on a single page, and other benefits while allowing me to continue Linux benchmarking and in cases like this purchasing hardware when necessary.