diff --git a/Grid/stencil/Stencil.h b/Grid/stencil/Stencil.h
index 766862fec..687b96219 100644
--- a/Grid/stencil/Stencil.h
+++ b/Grid/stencil/Stencil.h
@@ -315,6 +315,10 @@ public:
   }
 
   int              traceID;
+  double           OffNodeBytes;
+  double           CommTimer;
+  double           InterNodeBandwidthMBps;
+  
   int face_table_computed;
   //  int partialDirichlet;
   int fullDirichlet;
@@ -539,11 +543,12 @@ public:
 #endif
     }
     traceID = traceStart("Stencil::CommunicateBegin");
-    
+    OffNodeBytes=0;
+    CommTimer=-usecond();
     for(int i=0;i<Packets.size();i++){
       //      std::cout << "Communicate prepare "<<i<<std::endl;
       //      _grid->Barrier();
-      _grid->StencilSendToRecvFromPrepare(MpiReqs,
+      OffNodeBytes+=_grid->StencilSendToRecvFromPrepare(MpiReqs,
 					  Packets[i].compressed_send_buf,
 					  Packets[i].to_rank,Packets[i].do_send,
 					  Packets[i].compressed_recv_buf,
@@ -560,7 +565,7 @@ public:
     for(int i=0;i<Packets.size();i++){
       //      std::cout << "Communicate Begin "<<i<<std::endl;
       //      _grid->Barrier();
-      _grid->StencilSendToRecvFromBegin(MpiReqs,
+      OffNodeBytes+=_grid->StencilSendToRecvFromBegin(MpiReqs,
 					Packets[i].send_buf,Packets[i].compressed_send_buf,
 					Packets[i].to_rank,Packets[i].do_send,
 					Packets[i].recv_buf,Packets[i].compressed_recv_buf,
@@ -591,6 +596,8 @@ public:
     _grid->StencilSendToRecvFromComplete(MpiReqs,0); // MPI is done
     //    if   ( this->partialDirichlet ) DslashLogPartial();
     traceStop(traceID);
+    CommTimer+=usecond();
+    InterNodeBandwidthMBps = OffNodeBytes/CommTimer;
     
     if ( this->fullDirichlet ) DslashLogDirichlet();
     else DslashLogFull();
diff --git a/benchmarks/Benchmark_dwf.cc b/benchmarks/Benchmark_dwf.cc
index c28b2686a..2d21976f8 100644
--- a/benchmarks/Benchmark_dwf.cc
+++ b/benchmarks/Benchmark_dwf.cc
@@ -276,12 +276,14 @@ void Benchmark(int Ls, Coordinate Dirichlet,bool sloppy)
     Dw.Dhop(src,result,0);
     std::cout<<GridLogMessage<<"Called warmup"<<std::endl;
     double t0=usecond();
+    double aveBW = 0;
     for(int i=0;i<ncall;i++){
       Dw.Dhop(src,result,0);
+      aveBW +=Dw.Stencil.InterNodeBandwidthMBps/ncall;
     }
     double t1=usecond();
     FGrid->Barrier();
-
+    
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=single_site_flops*volume*ncall;
 
@@ -298,6 +300,7 @@ void Benchmark(int Ls, Coordinate Dirichlet,bool sloppy)
     std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
     std::cout<<GridLogMessage << "mflop/s per node =  "<< flops/(t1-t0)/NN<<std::endl;
+    std::cout<<GridLogMessage << "average bandwidth =  "<< aveBW <<std::endl;
     err = ref-result;
     n2e = norm2(err);
     std::cout<<GridLogMessage << "norm diff   "<< n2e<< "  Line "<<__LINE__ <<std::endl;