diff --git a/cime_config/allactive/config_pesall.xml b/cime_config/allactive/config_pesall.xml
index 39b155427184..b18dce68af8e 100644
--- a/cime_config/allactive/config_pesall.xml
+++ b/cime_config/allactive/config_pesall.xml
@@ -2548,4 +2548,80 @@
       </pes>
     </mach>
   </grid>
+  <grid name="a%ne256np4.pg2.*">
+    <mach name="pm-gpu">
+      <pes compset=".*SCREAM.*ELM.*MPASSI.*MPASO.*" pesize="S">
+        <comment>pm-gpu ne256 for fully coupled cases with ATM on GPU, MPAS on CPU -- WCYCLXX2010 64 nodes, 64x1</comment>
+        <MAX_MPITASKS_PER_NODE>64</MAX_MPITASKS_PER_NODE>
+        <ntasks>
+          <ntasks_atm>256</ntasks_atm>
+          <ntasks_cpl>-64</ntasks_cpl>
+          <ntasks_lnd>-64</ntasks_lnd>
+          <ntasks_rof>-64</ntasks_rof>
+          <ntasks_ice>-64</ntasks_ice>
+          <ntasks_ocn>-64</ntasks_ocn>
+        </ntasks>
+        <rootpe>
+          <rootpe_glc>1</rootpe_glc>
+          <rootpe_wav>1</rootpe_wav>
+          <rootpe_esp>1</rootpe_esp>
+          <rootpe_iac>1</rootpe_iac>
+        </rootpe>
+        <pstrid>
+          <pstrid_atm>16</pstrid_atm>
+        </pstrid>
+        <excl_stride>
+          <excl_stride_atm>16</excl_stride_atm>
+        </excl_stride>
+      </pes>
+      <pes compset=".*SCREAM.*ELM.*MPASSI.*MPASO.*" pesize="M">
+        <comment>pm-gpu ne256 for fully coupled cases with ATM on GPU, MPAS on CPU -- WCYCLXX2010 128 nodes, 64x1</comment>
+        <MAX_MPITASKS_PER_NODE>64</MAX_MPITASKS_PER_NODE>
+        <ntasks>
+          <ntasks_atm>512</ntasks_atm>
+          <ntasks_cpl>-128</ntasks_cpl>
+          <ntasks_lnd>-128</ntasks_lnd>
+          <ntasks_rof>-128</ntasks_rof>
+          <ntasks_ice>-128</ntasks_ice>
+          <ntasks_ocn>-128</ntasks_ocn>
+        </ntasks>
+        <rootpe>
+          <rootpe_glc>1</rootpe_glc>
+          <rootpe_wav>1</rootpe_wav>
+          <rootpe_esp>1</rootpe_esp>
+          <rootpe_iac>1</rootpe_iac>
+        </rootpe>
+        <pstrid>
+          <pstrid_atm>16</pstrid_atm>
+        </pstrid>
+        <excl_stride>
+          <excl_stride_atm>16</excl_stride_atm>
+        </excl_stride>
+      </pes>
+      <pes compset=".*SCREAM.*ELM.*MPASSI.*MPASO.*" pesize="M">
+        <comment>pm-gpu ne256 for fully coupled cases with ATM on GPU, MPAS on CPU -- WCYCLXX2010 128 nodes, 64x1</comment>
+        <MAX_MPITASKS_PER_NODE>64</MAX_MPITASKS_PER_NODE>
+        <ntasks>
+          <ntasks_atm>1024</ntasks_atm>
+          <ntasks_cpl>-256</ntasks_cpl>
+          <ntasks_lnd>-256</ntasks_lnd>
+          <ntasks_rof>-256</ntasks_rof>
+          <ntasks_ice>-256</ntasks_ice>
+          <ntasks_ocn>-256</ntasks_ocn>
+        </ntasks>
+        <rootpe>
+          <rootpe_glc>1</rootpe_glc>
+          <rootpe_wav>1</rootpe_wav>
+          <rootpe_esp>1</rootpe_esp>
+          <rootpe_iac>1</rootpe_iac>
+        </rootpe>
+        <pstrid>
+          <pstrid_atm>16</pstrid_atm>
+        </pstrid>
+        <excl_stride>
+          <excl_stride_atm>16</excl_stride_atm>
+        </excl_stride>
+      </pes>
+    </mach>
+  </grid>
 </config_pes>
diff --git a/cime_config/machines/config_batch.xml b/cime_config/machines/config_batch.xml
index f0cfe6800f8e..b1fbb5047ec3 100644
--- a/cime_config/machines/config_batch.xml
+++ b/cime_config/machines/config_batch.xml
@@ -411,15 +411,7 @@
     <directives>
       <directive> --constraint=gpu</directive>
     </directives>
-    <directives COMPSET="!.*MMF.*" compiler="gnugpu">
-      <directive> --gpus-per-node=4</directive>
-      <directive> --gpu-bind=none</directive>
-    </directives>
-    <directives COMPSET=".*MMF.*" compiler="gnugpu">
-      <directive> --gpus-per-task=1</directive>
-      <directive> --gpu-bind=map_gpu:0,1,2,3</directive>
-    </directives>
-    <directives compiler="nvidiagpu">
+    <directives compiler=".*gpu">
       <directive> --gpus-per-node=4</directive>
       <directive> --gpu-bind=none</directive>
     </directives>
diff --git a/cime_config/machines/config_machines.xml b/cime_config/machines/config_machines.xml
index cf1fbde6fd89..ba9b6f75f862 100644
--- a/cime_config/machines/config_machines.xml
+++ b/cime_config/machines/config_machines.xml
@@ -350,8 +350,8 @@
     <BATCH_SYSTEM>nersc_slurm</BATCH_SYSTEM>
     <SUPPORTED_BY>e3sm</SUPPORTED_BY>
     <MAX_TASKS_PER_NODE>128</MAX_TASKS_PER_NODE>
-    <MAX_TASKS_PER_NODE compiler="gnu">256</MAX_TASKS_PER_NODE>
-    <MAX_TASKS_PER_NODE compiler="nvidia">256</MAX_TASKS_PER_NODE>
+    <MAX_TASKS_PER_NODE compiler="gnu">128</MAX_TASKS_PER_NODE>
+    <MAX_TASKS_PER_NODE compiler="nvidia">128</MAX_TASKS_PER_NODE>
     <MAX_MPITASKS_PER_NODE>4</MAX_MPITASKS_PER_NODE>
     <MAX_MPITASKS_PER_NODE compiler="gnu">64</MAX_MPITASKS_PER_NODE>
     <MAX_MPITASKS_PER_NODE compiler="nvidia">64</MAX_MPITASKS_PER_NODE>
@@ -362,8 +362,9 @@
         <arg name="label"> --label</arg>
         <arg name="num_tasks"> -n {{ total_tasks }} -N {{ num_nodes }}</arg>
         <arg name="thread_count">-c $SHELL{echo 128/`./xmlquery --value MAX_MPITASKS_PER_NODE`|bc}</arg>
-        <arg name="binding"> $SHELL{if [ 64 -ge `./xmlquery --value MAX_MPITASKS_PER_NODE` ]; then echo "--cpu_bind=cores"; else echo "--cpu_bind=threads";fi;} </arg>
+        <arg name="binding"> $SHELL{if [ 64 -ge `./xmlquery --value MAX_MPITASKS_PER_NODE` ]; then echo "--cpu-bind=cores"; else echo "--cpu-bind=threads";fi;} </arg>
         <arg name="placement"> -m plane=$SHELL{echo `./xmlquery --value MAX_MPITASKS_PER_NODE`}</arg>
+        <arg name="gpu-bind"> /global/cfs/cdirs/e3sm/tools/set_affinity_npergpu.sh $SHELL{echo `./xmlquery --value MAX_MPITASKS_PER_NODE`}</arg>
     </arguments>
     </mpirun>
     <module_system type="module" allow_error="true">
@@ -453,9 +454,6 @@
       <env name="MPICH_ENV_DISPLAY">1</env>
       <env name="MPICH_VERSION_DISPLAY">1</env>
       <env name="MPICH_MPIIO_DVS_MAXNODES">1</env>
-      <env name="OMP_STACKSIZE">128M</env>
-      <env name="OMP_PROC_BIND">spread</env>
-      <env name="OMP_PLACES">threads</env>
       <env name="HDF5_USE_FILE_LOCKING">FALSE</env>
       <env name="PERL5LIB">/global/cfs/cdirs/e3sm/perl/lib/perl5-only-switch</env>
       <env name="FI_MR_CACHE_MONITOR">kdreg2</env>
@@ -463,6 +461,11 @@
       <env name="NETCDF_PATH">$ENV{CRAY_NETCDF_HDF5PARALLEL_PREFIX}</env>
       <env name="PNETCDF_PATH">$ENV{CRAY_PARALLEL_NETCDF_PREFIX}</env>
     </environment_variables>
+    <environment_variables BUILD_THREADED="TRUE">
+      <env name="OMP_STACKSIZE">128M</env>
+      <env name="OMP_PROC_BIND">spread</env>
+      <env name="OMP_PLACES">threads</env>
+    </environment_variables>
     <environment_variables compiler="gnugpu">
       <env name="MPICH_GPU_SUPPORT_ENABLED">1</env>
     </environment_variables>