diff --git a/rpd_tracer/Logger.cpp b/rpd_tracer/Logger.cpp
index b4e4975..f24a8b7 100644
--- a/rpd_tracer/Logger.cpp
+++ b/rpd_tracer/Logger.cpp
@@ -243,6 +243,17 @@ void Logger::init()
         if (autostart == 0)
             startTracing = false;
     }
+
+    // Add delay before starting tracing/autoflush if RPDT_DELAY is set
+    const char *delay_env = getenv("RPDT_DELAY");
+    if (delay_env != nullptr) {
+        int delay_sec = atoi(delay_env);
+        if (delay_sec > 0) {
+            fprintf(stderr, "rpd_tracer: delaying start by %d seconds (RPDT_DELAY)\n", delay_sec);
+            sleep(delay_sec);
+        }
+    }
+
     if (startTracing == true) {
         for (auto it = m_sources.begin(); it != m_sources.end(); ++it)
             (*it)->startTracing();
diff --git a/rpd_tracer/README.md b/rpd_tracer/README.md
index f1d1456..d70c1b6 100755
--- a/rpd_tracer/README.md
+++ b/rpd_tracer/README.md
@@ -61,3 +61,12 @@ This is a tracer that can attach to any process and record hip apis, ops, and ro
         init_tracing();
         ```
 <b>Note:</b> You can utilize `nm -gd <PATH_TO_librpd_tracer.so>` to find out symbol names in your library.
+
+<b>Delay Option:</b>
+- You can delay the start of profiling/autoflush by setting the environment variable `RPDT_DELAY` to the number of seconds to wait before tracing begins. This is useful for large workloads where you want to skip initial setup time.
+- Example usage:
+  ```bash
+  export RPDT_DELAY=10  # Delay start by 10 seconds
+  export RPDT_AUTOFLUSH=1
+  <your launch command>
+  ```