CUDA, Supercomputing for the Masses: Part 20

CUDA, Supercomputing for the Masses: Part 19, I focused on the new Parallel Nsight tool for Microsoft Visual Studio including: the thought that went into the design of the tool, how to install and configure the software, how to create a project from scratch, and provided a demonstration of the debugging capabilities provided in the 1.0 release. This article utilizes the Parallel Nsight 1.0 analysis capabilities coupled with the NVIDIA Tools Extension (NVTX) library to examine the simpleMultiCopy asynchronous I/O example from the NVIDIA GPU Computing SDK; create a hybrid CPU and GPU computation that balances the simultaneous use of both CPU and GPU resources on a single calculation; and compare the very fast primitive restart OpenGL rendering code from Part 18 with more conventional OpenGL rendering methods. If you're interested in any of these topics and thinking about purchasing the professional version of Parallel Nsight, you should find this article illuminating. The professional version of Parallel Nsight provides the ability to trace and analyze:

• CUDA applications. In particular, Parallel Nsight has the ability to insert code to capture application traces that include operating system calls, CUDA kernels, data transfers, and host-based methods. NVIDIA put this video online to give viewers a sense of the Nsight tracing and analysis capabilities.
• OpenCL applications. Parallel Nsight has been designed to support multiple languages so OpenCL applications can be traced with the same functionality as CUDA applications.
• Graphics Analysis. For game and visual computing developers, Parallel Nsight allows for the tracing of OpenGL and DirectX API calls in version 1.0, with the tracing of graphics GPU workloads coming in the upcoming version 1.5. This is in addition to the Graphics Inspector, which is part of Parallel Nsight Standard. The Graphics Inspector provides a heads-up performance display (HUD), real-time profiling, plus the ability to examine performance markers and inspect the graphics pipelines. While not covered in this article, NVIDIA demonstrates these capabilities and more in this video. Neither Version 1.0 or 1.5 support OpenGL in the Graphics Inspector.
• Note that the Graphics Inspector is part of the standard version.
• Version 1.5 of Parallel Nsight adds a Graphics Workload Trace to the analysis capability, which traces the GPU workloads spawned by individual draw calls.

Information Sources about Parallel Nsight

As mentioned in the previous article, the following are excellent sources of information about Parallel Nsight:

• The NVIDIA Parallel Nsight forum is an excellent place to look for information and post questions.
• The Parallel Nsight Developer Zone is the main entry point for Parallel Nsight on the NVIDIA site. It provides links to support, videos and webinars.