From 60dcf7d0a8de4e08f38aec92e13932f8cf88f145 Mon Sep 17 00:00:00 2001 From: Nicole Date: Tue, 14 Feb 2023 15:01:32 -0800 Subject: [PATCH] Add rocm-smi to GPU docs (#1102) * Add rocm-smi to gpu docs * spellcheck sampling --------- Co-authored-by: Nicole Marsaglia --- src/docs/sphinx/Actions/Binning.rst | 6 +++--- src/docs/sphinx/GPU_notes.rst | 13 +++++++++++++ 2 files changed, 16 insertions(+), 3 deletions(-) diff --git a/src/docs/sphinx/Actions/Binning.rst b/src/docs/sphinx/Actions/Binning.rst index 8c02f7e9a..9c682a9a0 100644 --- a/src/docs/sphinx/Actions/Binning.rst +++ b/src/docs/sphinx/Actions/Binning.rst @@ -35,8 +35,8 @@ simulation codes, regardless of the underlying mesh representation. Sampling and Aggregation ------------------------ -When specifying the number of bins on an axis, there will always be over smapling or undersampling. -During spatial binning, each zone is placed into a bin based on it centriod, and as with all +When specifying the number of bins on an axis, there will always be over sampling or undersampling. +During spatial binning, each zone is placed into a bin based on its centriod, and as with all binning, this is subject to over sampling or under sampling. @@ -62,7 +62,7 @@ example. An example of spatial over sampling. -When oversamping data, the default value of an empy bin is 0. That said, the default empty +When oversampling data, the default value of an empty bin is 0. That said, the default empty value can be overridden by an option named parameter, e.g., `empty_bin_val=100`. This is often useful when the default value is part of the data range, and setting the empty bin value to something known, allows the user to filter out empty bins diff --git a/src/docs/sphinx/GPU_notes.rst b/src/docs/sphinx/GPU_notes.rst index 70a7e5d10..69d02a7b2 100644 --- a/src/docs/sphinx/GPU_notes.rst +++ b/src/docs/sphinx/GPU_notes.rst @@ -16,6 +16,19 @@ By default, each CUDA capable GPU device is queried, and a rank is assigned a de Collisions could result in a run-time failure or significant delays. This default behavior can be overridden. Please see the Ascent options for more details. +Using ROCm-SMI on AMD GPUs +-------------------------- +ROCm-SMI stands for ROCm System Management Interface and provides an interfaces for applications and users to monitor and control GPU applicatoins. + +ROCm-SMI can be used to verify GPU usage. +This is done by launching your application and then using ROCm-SMI on the same node to monitor the running job. + +The following call will monitor the usage of GPUs every tenth of a second: + +.. code-block:: sh + + watch -n 0.1 rocm-smi + Using RocProf on AMD GPUs ------------------------- RocProf is a command line tool provided by ROCm that is implemented on top of the rocProfiler and rocTracer APIs.