Force ring algorithm on single node (#509)

src/graph/tuning.cc

@@ -126,7 +126,7 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
       nNodes;
 
     for (int a=0; a<NCCL_NUM_ALGORITHMS; a++) {
-      if (coll != ncclFuncAllReduce && a != NCCL_ALGO_RING) continue;
+      if ((coll != ncclFuncAllReduce || nNodes == 1) && a != NCCL_ALGO_RING) continue;
 
       for (int p=0; p<NCCL_NUM_PROTOCOLS; p++) {
         float speed = nNodes <= 2 || a == NCCL_ALGO_COLLNET ? graphs[a]->speedIntra : graphs[a]->speedInter;