Multi-core programming with x86 CPUs is difficult and often results in marginal performance gain when going from 1 core to 4 cores to 16 cores. Beyond 4 cores, memory bandwidth becomes the bottleneck to further performance increases.

To harness the parallel computing power of GPUs, programmers can simply modify the performancecritical portions of an application to take advantage of the hundreds of parallel cores in the GPU. The rest of the application remains the same, making the most efficient use of all cores in the system. Running a function of the GPU involves rewriting that function to expose its parallelism, then adding a fewnew function-calls to indicate which functions will run on the GPU or the CPU. With these modifications, the performance-critical portions of the application can now run significantly faster on the GPU.

CUDA™ is NVIDIA’s parallel computing architecture. Applications that leverage the CUDA architecture can be developed in a variety of languages and APIs, including C, C++, Fortran, OpenCL, and DirectCompute. The CUDA architecture contains hundreds of cores capable of running many thousands of parallel threads, while the CUDA programming model lets programmers focus on parallelizing their algorithms and not the mechanics of the language.

The latest generation CUDA architecture, codenamed “Fermi”, is the most advanced GPU computing architecture ever built. With over three billion transistors, Fermi is making GPU and CPU co-processing pervasive by addressing the fullspectrum of computing applications.

With support for C++, GPUs based on the Fermi architecture make parallel processing easier and accelerate performance on a wider array of applications than ever before. Just a few applications that can experience significant performance benefits include ray tracing, finite element analysis, high-precision scientific computing, sparse linear algebra, sorting, and search algorithms.