Cosmetique.

README.md

@@ -189,9 +189,11 @@ the kernel code. The SYCL implementation will take in charge the move of the dat
 host memory and the device memory, as needed.
 
 This is very simple to use. Yet, since the developer does provide any information about what
-data will be read and what data will ne written (on the contrary of buffers and accessors),
-we can expect that the performance may be poorer for some use-cases. Try on your `MyProject` code, and 
-make up your own mind.
+data will be read and what data will be written (on the contrary of buffers and accessors),
+we can expect that the performance may be poorer for some use-cases.
+
+Try on your `MyProject` code, and  make up your own mind. So to work on very I/O intensive
+jobs, we suggest to ask for 500 images of 1 iterations.
 
 WARNING: all those I/O considerations really make sense if your are running on a discrete
 GPU, and the device has a dedicated memory, separated from the host one.