diff --git a/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc b/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc
index 167f24796593c4c643a52c2bb6c95feffaf759d9..2a6a03389f4356e0f1e15d6266eb5e0be61788d4 100644
--- a/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc
+++ b/paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc
@@ -813,12 +813,6 @@ void BuildOpFuncList(const platform::Place& place,
 
     interpreter::LogDeviceMemoryStats(place);
   }
-
-  // NOTE(Ruibiao): Release memory cache to avoid memory fragments in Allocator.
-  // It reduce about 10% memory usage for V100 8-GPU training of
-  // transformer_base_bs4096_amp_fp16 and transformer_base_bs4096_pure_fp16
-  // model.
-  memory::Release(place);
 }
 
 void LogDeviceMemoryStats(const platform::Place& place) {