diff --git a/tensorrt_llm/_torch/attention_backend/flashinfer.py b/tensorrt_llm/_torch/attention_backend/flashinfer.py
index 49964319a18a..3322422075dc 100644
--- a/tensorrt_llm/_torch/attention_backend/flashinfer.py
+++ b/tensorrt_llm/_torch/attention_backend/flashinfer.py
@@ -1006,7 +1006,6 @@ def _plan_with_params(self,
                         self.kv_layout,
                         backend="cudnn",
                     ))
-            torch.cuda.current_stream().synchronize()
             if self.num_contexts <= 0:
                 raise ValueError(
                     "FlashInfer ragged prefill without KV cache requires "