New submitter shervinemami
writes (starting with a pretty big disclaimer: "I'm an Engineer at NVIDIA.") The latest CompuBench GPU benchmarks show NVIDIA's Tegra K1 running whole OpenCL algorithms around 5x faster than any other mobile device, and individual instructions around 20x faster! This huge jump is because mobile companies have been saying they support OpenCL on mobile devices since early 2013, but what they don't mention is that they only have software API support, not hardware-accelerated OpenCL running faster on their GPUs than CPUs. Now that NVIDIA's Tegra-K1 chip has started shipping in devices and thus is available for full benchmarking, it is clearly the only mobile chip that actually gives you proper hardware-accelerated OpenCL (and CUDA of course!).
The K1 is also what's in Google's Project Tango 3-D mapping tablet