[Isambard] Move memory to its own resource for Nvidia Ampere partition

SYSTEMS.md

@@ -115,6 +115,12 @@ reframe -c examples/sombrero -r --performance-report --system isambard-macs:rome
 
 You may also need to compile GPU applications on the compute nodes, as the login node does not have any GPUs (this really depends on the build system of the application at hand, whether it needs access to a GPU during the build or it is sufficient to have the GPU toolkit available).
 
+### Phase3 partition
+
+On the nodes with Nvidia Ampere GPUs, memory usage is restricted to 4 GiB by default.
+To request more memory you [have to specify the resource](https://gw4-isambard.github.io/docs/user-guide/PHASE3.html#nvidia-gpu) `mem=...G`, for example by setting the `memory` key of `extra_resources` to the amount of memory you require.
+This can be done on the command line of ReFrame [...]
+
 ## Myriad
 
 ### Python3 module

benchmarks/reframe_config.py

@@ -364,9 +364,12 @@ def spack_root_to_path():
                     'resources': [
                         {
                             'name': 'gpu',
-                             # TODO: memory should be a separate resource.
-                            'options': ['ngpus={num_gpus_per_node}:mem=20G'],
+                            'options': ['ngpus={num_gpus_per_node}'],
                         },
+                        {
+                            'name': 'memory',
+                            'options': ['mem={memory}'],
+                        }
                     ],
                 },
                 {