OpenCL: Single-Precision
Once more, we are purposely choosing an alternate set of benchmarks that allows us to compare single- and double-precision math. Our main question today is whether more memory bandwidth and a higher core clock allows GeForce GTX 760 to outpace the GTX 660 Ti and its greater number of CUDA cores.
Financial Analysis Performance (Float/FP32)
The older GeForce GTX 660 Ti swoops in for an easy win ahead of the GeForce GTX 670 and new GTX 760. In this sort of workload, frequency and shaders count more than memory bandwidth.
Folding@Home (Float/FP32)
The Folding@Home benchmark paints a similar picture, with the GeForce GTX 760 bringing up the rear, since its wider memory interface isn’t enough to offset a lower number of shader units.
The GeForce cards come out ahead in the Folding@Home benchmark, while the Radeons remain favorites in financial analysis. Looking specifically at the GeForce GTX 760, we can see that its lower number of compute resources hurts it in OpenCL and compute-heavy scenarios. The card was able to mask this shortcoming in gaming tests thanks to higher memory bandwidth, but when it comes to crunching numbers, that’s not a bottleneck.