diff --git a/CHANGELOG.md b/CHANGELOG.md
index a5891d6f..cb5be2aa 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,7 +5,7 @@
 - PIESNO will now warn if less than 1% of noisy voxels were identified, which might indicate that something has gone wrong during the noise estimation.
 - On python >= 3.4, --mp_method [a_valid_start_method](https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods) can now be used to control behavior in the multiprocessing loop.
 - A new option --split_b0s can be specified to split the b0s equally amongst the training data.
-- A new (kind of experimental) option --use_f32 can be specified to use the float32 mode of spams.
+- A new (kind of experimental) option --use_f32 can be specified to use the float32 mode of spams and reduce ram usage.
 - A new option --use_threading can be specified to disable python multiprocessing and solely rely on threading capabilities of the linear algebra libs during denoising.
 - Fixed crash in option --noise_est local_std when --cores 1 was also supplied.
 - setup.py and requirements.txt will now fetch spams v2.6, with patches for numpy 1.12 support.
diff --git a/nlsam/denoiser.py b/nlsam/denoiser.py
index b2dd526a..91295081 100644
--- a/nlsam/denoiser.py
+++ b/nlsam/denoiser.py
@@ -71,13 +71,13 @@ def nlsam_denoise(data, sigma, bvals, bvecs, block_size,
     b0_threshold : int, default 10
         A b-value below b0_threshold will be considered as a b0 image.
     dtype : np.float32 or np.float64, default np.float64
-        Precision to use for inner computation. Note that np.float32 should only be used for
-        very, very large datasets (that is, you ram starts swappping) as it can lead to numerical precision errors.
+        Precision to use for inner computations. Note that np.float32 should only be used for
+        very, very large datasets (that is, your ram starts swappping) as it can lead to numerical precision errors.
     use_threading : bool, default False
         Do not use multiprocessing, but rather rely on the multithreading capabilities of your numerical solvers.
-        While this mode is more memory friendly, it is undoubtedly slower than using the multiprocessing mode (the default).
+        While this mode is more memory friendly, it is also slower than using the multiprocessing mode (the default).
         Moreover, it also assumes that your blas/lapack/spams library are built with multithreading, so be sure to check
-        the resources usage of your computer to make sure it is the case or the algorithm will just take much longer to complete.
+        that your computer is using multiple cores or the algorithm will just take much longer to complete.
     verbose : bool, default False
         print useful messages.
     mp_method : string