diff --git a/docs/examples/10_inverse_rendering/forward_diff.py b/docs/examples/10_inverse_rendering/forward_diff.py
index 54082f2f4..af26d8e78 100644
--- a/docs/examples/10_inverse_rendering/forward_diff.py
+++ b/docs/examples/10_inverse_rendering/forward_diff.py
@@ -35,4 +35,6 @@
 
 # .. write them to a PNG file
 crop_size = scene.sensors()[0].film().crop_size()
-write_bitmap('out.png', image_grad, crop_size)
+fname = 'out.png'
+write_bitmap(fname, image_grad, crop_size)
+print('Wrote forward differentiation image to: {}'.format(fname))
diff --git a/docs/src/getting_started/faq.rst b/docs/src/getting_started/faq.rst
index 7847f6165..ee25e55a5 100644
--- a/docs/src/getting_started/faq.rst
+++ b/docs/src/getting_started/faq.rst
@@ -27,6 +27,13 @@ Frequently asked questions
     compatible with the video driver (the two must satisfy version requirements
     that are detailed on the OptiX website)
 
+- Differentiable rendering fails with an error message similar to ":monosp:`RuntimeError:
+  cuda_malloc(): out of memory!`".
+
+    The rendering or autodiff graph exhausted your GPU's memory. Try reducing the sample count
+    and / or resolution. If high sample count renders are necessary, it is possible to split computation
+    in several low sample count passes and average together the gradients.
+
 - Citing Mitsuba 2 in scientific literature.
 
   Please use the following BibTeX entry to cite Mitsuba 2 in research articles,
@@ -45,7 +52,7 @@ Frequently asked questions
           doi = {10.1145/3355089.3356498}
       }
 
-  Here is a plain-text version of the above: 
+  Here is a plain-text version of the above:
 
       Merlin Nimier-David, Delio Vicini, Tizian Zeltner, and Wenzel Jakob.
       2019. Mitsuba 2: A Retargetable Forward and Inverse Renderer. In
diff --git a/docs/src/inverse_rendering/diff_render.rst b/docs/src/inverse_rendering/diff_render.rst
index 6ff68b59e..9f7245624 100644
--- a/docs/src/inverse_rendering/diff_render.rst
+++ b/docs/src/inverse_rendering/diff_render.rst
@@ -11,7 +11,8 @@ well-known Cornell Box scene that can be downloaded `here
 Please make the following three changes to the ``cbox.xml`` file:
 
 1. ``ldsampler`` must be replaced by ``independent`` (``ldsampler`` has not yet
-   been ported to Mitsuba 2)
+   been ported to Mitsuba 2). Note that the sample count specified here will be overriden
+   in the optimization script below.
 
 2. The integrator at the top should be defined as follows: