Nsight Visual Studio Edition 3.0 New Features

This new release officially supports OpenGL frame debugging and profiling, GLSL GPU shader debugging, local single GPU shader debugging, the new Kepler? GK110 architecture found in Tesla? K20 & GeForce GTX TITAN, and CUDA? 5.0.

For a complete overview of all Nsight? Visual Studio Edition features and access to resources, please visit the Nsight? Visual Studio Edition 3.0 page.

Register for free access to latest Nsight? Visual Studio Edition releases.

Download Documentation

New NVIDIA? Nsight? Visual Studio Edition 3.0 Features:

Graphics Debugging and Profiling

• Support OpenGL 4.2 Head-Up-Display to overlay key performance statistics.
• Support OpenGL 4.2 for frame debugging, pixel history and frame profiling.
  Note: The OpenGL application must be 4.2 core compliant in order to use local shader debugging.
• Support for OpenGL GLSL GPU shader debugging.
  • GLSL 3.3 and higher applications are supported on a remote debugging setup.
  • GLSL 4.2 core applications are supported on a local debugging setup.
• Local, single GPU shader debugging and pixel history is now supported for HLSL and GLSL.
• DirectX frame capture generates Visual Studio project with source code.
• Improved frame debugger Visual Studio frame scrubbing performance.

Compute Debugging

• Support for the CUDA 5.0? Toolkit.
• Support for the Kepler GK110 architecture (for example, found in the Tesla? K20).
• CUDA Dynamic Parallelism is now supported when building, debugging, and running analysis. For more information, see CUDA Dynamic Parallelism.
• The CUDA memory checker now supports Kepler GK110 and CUDA Dynamic Parallelism. For more information, see CUDA Dynamic Parallelism.
• Attaching to a CUDA application during a kernel launch is now supported. Attachable programs will raise an attach dialog if a GPU assert or exception occurs.
• The CUDA Information tool window now displays information about the application's CUDA textures and surfaces.
• Concurrent grid launches are now supported under the CUDA Debugger on all Kepler GPUs.
• Expressions can be moved to and from the Warp Watch window via drag-and-drop.
• Support for hardware debugging of C++ AMP applications directly on NVIDIA hardware.

Figure 1: Attach dialog when the CUDA application has hit a GPU assert or exception.

System Trace and Compute Profiling

• Support for the CUDA 5.0? Toolkit.
• Support for the Kepler GK110 architecture (for example, found in the Tesla? K20).
• Trace CUDA Dynamic Parallelism is now supported. For more information, see CUDA Dynamic Parallelism.
• New CUDA Source-Level Experiments allow you to correlate profiling data to individual kernel source code lines, supporting CUDA-C, PTX, and SASS assembly.
• CUDA Queue Trace provides insight to the state and depth of the CPU/GPU workload queues.
• CUDA Active Warp Time Trace reports the sum of all warp execution durations per kernel launch.
• Ability to augment the NVIDIA Nsight Analysis timeline with custom data through the new NVTXT file type.
• NVTX and DX events now support collecting stack traces, as well as flat sorting for its ranges.
• Added tracing of the Windows Display Driver Model (WDDM) to collect data of the GPU utilization, context queues, hardware queues, and GPU contexts switches.

Figure 2: With CUDA Source-Level Experiments, you can correlate profiling data to individual kernel source code lines