Introduction Wen-mei W. Hwu STATE OF GPU COMPUTING We are entering the golden age of GPU computing. Since the introduction of CUDA in 2007, more than 100 million computers with CUDA-capable GPUs have been shipped to end users. Unlike the previous GPGPU shader programming models, CUDA supports parallel programming in C. From my own experience in teaching CUDA programming, C programmers can begin to write basic CUDA programs after only attending one lecture and reading one textbook chapter. With such a low barrier of entry, researchers all over the world have been engaged in developing new algorithms and applications to take advantage of the extreme floating point execution throughout these GPUs. Today, there is a large community of GPU computing practitioners. Many of them have reported a 10 to 100 times speedup of their applications with GPU computing. To put this into perspective, with the historical 2X performance growth every 2 years, these researchers are experiencing the equivalent of time travel of 8 to 12 years. That is, they are getting the performance today that they would have to wait for 8 to 12 years if they went for the "free-ride" advancement of performance in microprocessors.