diff --git a/sphinx_docs/source/olcf-compilers.rst b/sphinx_docs/source/olcf-compilers.rst
index c6353ed..fe94104 100644
--- a/sphinx_docs/source/olcf-compilers.rst
+++ b/sphinx_docs/source/olcf-compilers.rst
@@ -16,13 +16,13 @@ Then load CUDA::
 
 You also need to make sure you have the python module loaded::
 
-  module load python/3.5.2
+  module load python/3.7.0
 
 Compile with ``COMP = pgi`` and ``USE_CUDA=TRUE``.  Ensure your
 GNUMakefile uses ``USE_OMP=FALSE`` since AMReX's standard OpenMP
 strategy conflicts with GPUs.  An example compilation line is::
 
-  make COMP=pgi INTEGRATOR_DIR=VODE90 USE_CUDA=TRUE -j 4
+  make COMP=pgi INTEGRATOR_DIR=VODE USE_CUDA=TRUE -j 4
 
 
 The version pairs that work for sure currently are:
@@ -40,9 +40,6 @@ The version pairs that work for sure currently are:
 
 .. note::
 
-   - Use ``INTEGRATOR_DIR=VODE90`` if specifying an integrator at
-     compile time, since ``VODE`` will not work with GPUs.
-
    - Use ``USE_GPU_PRAGMA=TRUE`` for any code which uses ``#pragma
      gpu``, this flag is set by default in MAESTROeX and CASTRO's
      build system when ``USE_CUDA=TRUE``.
@@ -50,4 +47,13 @@ The version pairs that work for sure currently are:
    - OpenMP offloading to the GPU is controlled by
      ``USE_OMP_OFFLOAD``, which will default to ``FALSE``, AMReX-Astro
      doesn't use this feature.
+   
+   - MAESTROeX requires C++14, so it is necessary to use a   
+     later version of gcc than the version loaded on Summit 
+     when the pgi module is loaded. To do this, add the 
+     following to the ``GNUMakefile``::
+
+       NVCC_CCBIN=/sw/summit/gcc/7.4.0/bin/g++
+       LIBRARY_LOCATIONS += /sw/summit/gcc/7.4.0/lib64/
+       LIBRARIES += -Wl,-rpath,/sw/summit/gcc/7.4.0/lib64/