High-Performance MATLAB with GPU Acceleration
18 Aug 2015 - Joss Knight discusses on a NVIDIA blog several techniques one can use to maximize the performance of GPU-accelerated MATLAB code. In conclusion: "Vectorization is a key concept for MATLAB programming that provides big performance advantages for both standard and GPU coding. Combining this principle with the wealth of built-in functions optimized for the GPU is usually enough to take advantage of ones device, without needing to learn about CUDA or parallel programming concepts. And if that falls short, the arrayfun function lets you craft custom kernels in the MATLAB language to eke a lot more performance from ones card", says Knight.