Fix image display benchmark notebook.

docs/benchmarking.ipynb

@@ -8,7 +8,7 @@
    "source": [
     "# Benchmarks\n",
     "\n",
-    "A demonstration of parallel processing performance of this CNGI prototype against the current released version of CASA using a selection of ALMA datasets representing different computationally demanding configurations and a subset of VLA CHILES Survey data."
+    "A demonstration of parallel processing performance of this CNGI prototype against the current released version of CASA using a selection of ALMA datasets representing different computationally demanding configurations and a subset of the VLA CHILES Survey data."
    ]
   },
   {
@@ -158,11 +158,11 @@
     "\n",
     "Neither CASA nor the prototype achieves a linear speedup due to scheduler overhead, disk io, internode communication, etc. The most computationally expensive step is the gridding of the convolved visibilities onto the uv grid:\n",
     "\n",
-    "![im340](https://raw.githubusercontent.com/casangi/cngi_prototype/master/docs/_media/gcf_size.png)\n",
+    "![im990](https://raw.githubusercontent.com/casangi/cngi_prototype/master/docs/_media/gcf_size.png)\n",
     "\n",
     "Since the gridding step can be done in parallel, increasing gridding convolution support will increase the part of the work that can be done in parallel. In the figure below, the support size of the gridding convolution kernel is increased from 17x17 to 51x51, consequently, a more linear speedup is achieved.\n",
     "\n",
-    "![im340](https://raw.githubusercontent.com/casangi/cngi_prototype/master/docs/_media/gcf_size_cluster_A001_X12a3_X3be.png)\n",
+    "![im998](https://raw.githubusercontent.com/casangi/cngi_prototype/master/docs/_media/gcf_size_cluster_A001_X12a3_X3be.png)\n",
     "\n"
    ]
   },