OpenCL: GPGPU Benchmarks
In addition to including the current crop of workstation cards from both companies, we have the most important models from last generation. Depending on the OpenCL-enabled application, workload, and degree of driver optimization, we end up with varied results. Yet, we also see a distinct trend.
AMD’s workstation and consumer cards take a very clear lead in bitmining and SHA-256 hashing. If bitcoins are what you seek, the GeForce GTX 660 is not for you. In fact, the previous-gen GeForce GTX 580, sporting GF110, turns out to be the fastest Nvidia card in this field, serving as a reminder of how deliberately pared-down the compute resources of Kepler-based desktop cards really are.
We purposely chose LuxMark 2.0's simplest rendering workload for this test. Otherwise, Nvidia’s lower-end models would have given us even shorter bars, making them indiscernible. The previous-generation’s Fermi-based GeForce GTX 570 and 580 enjoy more pronounced leads as the scene's complexity increases. Once more, AMD’s entire line-up is completely out of any GeForce card’s reach.
The PostFX test utilizes both OpenCL and OpenGL to create one demanding workload. Although this benchmark comes straight from Nvidia’s SDK, it’s also an impressive example of how quickly the performance story can change in the face of a new architecture like GCN.
At last, Nvidia's Kepler family exhibits a stronger showing. Nonetheless, it's the company's previous-gen architecture that comes out on top once more. We're taken aback by the OEM version’s good results, since it even outpaces the GeForce GTX 660 Ti. Worried that this was a testing issue, we re-ran the benchmark several times and came up with consistent scores.