2.7.3-1

Add support for A100 GPU and related platforms.
Add support for CUDA 11.
Add support for send/receive operations (beta).

src/graph/xml.cc

@@ -590,7 +590,7 @@ ncclResult_t ncclTopoGetXmlFromGpu(struct ncclXmlNode* pciNode, nvmlDevice_t nvm
   NCCLCHECK(xmlGetSub(pciNode, "nvlink", &nvlNode));
   if (nvlNode == NULL) {
     // NVML NVLink detection
-    int maxNvLinks = (sm < 60) ? 0 : (sm < 70) ? 4 : 6;
+    int maxNvLinks = (sm < 60) ? 0 : (sm < 70) ? 4 : (sm < 80) ? 6 : 12;
 
     if (maxNvLinks > 0 && nvmlDev == NULL) {
       WARN("No NVML device handle. Skipping nvlink detection.\n");