Nsight Compute 2020.2 - New Features

New Application Replay complements the Kernel Replay activity
(See the Super Computing 2020 spotlight on these replay modes)

Fixed several issues related to auto-profiling in the UI.
Fixed a metric collection issue when profiling kernels on different GPU architectures with application replay.
Fixed a performance problem related to profiling large process trees.
Fixed that occupancy charts would not render correctly when comparing against baselines.
Fixed that no memory metrics were shown on the Source page for LDGSTS instructions.
Fixed the automatic sorting on the Summary and Raw pages.
Fixed an issue that would cause the NVIDIA Nsight Compute CLI to consume too much memory when importing or printing reports.
Long kernel names are now elided in the Details page source hot spot tables.
Fixed that function names in the Resources tool window were demangled differently.

Added support for the NVIDIA Ampere GA102 and GA104 GPU architectures.
Support for CUDA Toolkit 11.1 was added.
Added support for application replay (documentation, spotlight video) to collect metric results across multiple application runs, instead of replaying individual kernels.
Added new launch__device_id metric.

CLI output now shows NVTX payload information.
CSV output now shows NVTX states.
Added a new --replay-mode option to select the mechanism used for replaying a kernel launch multiple times.
Added a new --kill option to terminate the application once all requested kernels were profiled.
Added a new --log-file option to decide the output stream for printing tool output.
Added a new --check-exit-code option to decide if the child application exit code should be checked.

For a complete overview of all NVIDIA Nsight™ Compute features and access to resources, please visit the main Nsight™ Compute page.

NVIDIA® Nsight™ Compute 2020.2 is available for download under the NVIDIA Registered Developer Program.

Videos