CUDA and OpenCL Performance
First we checked out our graphics cards in SVPMark 3.0.3a from SmoothVideo Project (the real-life test):
We have almost the same score irrespective of the graphics card. This test just can’t use up all of the computing potential of these products.
Next we benchmarked the speed of HD video encoding in Xilisoft Video Converter version 7.7.2, which supports both AMD and Nvidia GPUs. During the test we converted a Full-HD 2.5GB video file into an iPad 4 format. The resulting file was 750 MB large. A smaller result is better here:
Interestingly, the standard GeForce GTX 770 turns out to be a little faster than the GeForce GTX 780, which must be due to the higher GPU clock rate because the GeForce GTX 680 performs well here, too. The Radeon HD 7970 GHz Edition encodes video at a slower rate than the Nvidia-based cards, yet is anyway faster than the overclocked six-core Intel CPU.
The next test is DirectCompute Benchmark version 0.45b. The diagram shows the DirectCompute and OpenCL scores:
AMD is unrivalled in DirectCompute calculations but the Nvidia-based cards are ahead of the Radeon HD 7970 GHz Edition in OpenCL.
The popular LuxMark benchmark produced rather interesting results:
The GeForce GTX 780’s GK110 chip has almost twice the computing capabilities of the GeForce GTX 770’s GK104 processor. The Radeon's Tahiti XT is far ahead of the opponents in LuxMark, though.
Here is a table with full test results: