NVIDIA Nsight Visual Studio Edition 2.2 User Guide

NVIDIA® Nsight™ Development Platform, Visual Studio Edition 2.2 User Guide
Send Feedback

NVIDIA® Nsight™ Development Platform, Visual Studio Edition is a development environment for CUDA and graphics applications running on NVIDIA GPUs, which is integrated into Microsoft Visual Studio.

With NVIDIA Nsight 2.2, you can use the NVIDIA Nsight CUDA™ Debugger, Graphics Debugger, and Analysis and Profiling Tools in either Visual Studio 2008 or Visual Studio 2010.

The NVIDIA Nsight tools extend the debugging and performance analysis capabilities of Visual Studio to support GPU computing. NVIDIA Nsight is useful in several different application areas, including:

• Game development,
• High-performance computing and supercomputing, and
• Workstation and content creation software.

CUDA Debugger

The CUDA Debugger helps you debug applications that use the Compute Unified Device Architecture (CUDA). You can set breakpoints in CUDA source code, inspect memory, view the values of local variables, perform memory checks, as well as other common debugging tasks. You can use the CUDA Debugger with applications built with the CUDA Runtime (CUDART) API or with the CUDA Driver API.

NVIDIA® Nsight™ Visual Studio Edition 2.2 introduces:

• Native hardware CUDA debugging on single GPU systems.
• Support for the CUDA 4.2 Toolkit.
• Support for the new Kepler architecture (for example, found in the GeForce GTX 680).
• The CUDA information view now shows the number and size of memory allocations per context per device.
• Change focus to a specific thread within a warp in the lanes column of the CUDA information view.
• New "freeze" and "thaw" commands to control warp execution in the CUDA information view.
• Drastic performance improvement when scrolling the warp page in the CUDA information view.
• Scroll through the disassembly in the entire address space.
• Ability to enter a function or symbol name in the disassembly toolbar.
• Support for creating a function breakpoint by using the stock Visual Studio breakpoint option.
• Ability to set a breakpoint from the Functions page in the CUDA information view.

Graphics Debugger

The Graphics Debugger allows you to debug frame by each draw call. You can debug vertex shaders, pixel shaders, and view pipeline states. You can create performance markers and profile the execution of your graphics code.

NVIDIA® Nsight™ Visual Studio Edition 2.2 introduces:

• Support for the new Kepler architecture (for example, found in the GeForce GTX 680).
• Support DirectX 9 for Frame Debugging, Frame Profiling, and analysis.
• Improved Frame Profiler with performance improvements and additional experiments.
• Improved Frame Debugger Visual Studio frame scrubbing performance.

Analysis and Profiling Tools

The Analysis tools help you to understand how workloads are distributed across your application. You can see API calls (including CUDA C/C++, OpenCL, DirectX, Microsoft DirectCompute, OpenGL, and Cg), memory copies, kernel executions, and CPU core and thread events along a visual timeline.

NVIDIA® Nsight™ Visual Studio Edition 2.2 introduces:

• Support for the new Kepler architecture (for example, found in the GeForce GTX 680).
• Support for the CUDA 4.2 Toolkit.
• Trace Activities

  • Add support for DirectX 9 Trace.

  • DirectX 10/11 Trace adds support for capturing memory transfers, shader compile times, and SLI events.
• Profile Activities

  • Significantly reduced performance overhead for kernel profiling.

  • Ability to skip the first N kernels that will be executed and then profile the next X kernels.

  • Profile CUDA Process tree collects performance data for the target application and all native child processes.
• Report System
  • Analysis summary pages have been redesigned to give a true high-level picture of how the application utilizes different areas.
  • Background bars in the analysis reports tables quickly shows critical values.
  • Usability of hierarchical report pages (NVTX page, DX Performance Markers) has been improved.
  • Added new render modes for the timeline axis (relative to capture, relative to cursor, and relative to viewport) to improve overall usability.
  • Reduced time to open report files, including system module information.
  • Optimized memory overhead and load time of the timeline for application traces.