diff --git a/superbench/benchmarks/micro_benchmarks/cpu_copy_performance/cpu_copy.cpp b/superbench/benchmarks/micro_benchmarks/cpu_copy_performance/cpu_copy.cpp
index 6831f4093..5f1193918 100644
--- a/superbench/benchmarks/micro_benchmarks/cpu_copy_performance/cpu_copy.cpp
+++ b/superbench/benchmarks/micro_benchmarks/cpu_copy_performance/cpu_copy.cpp
@@ -243,19 +243,18 @@ double BenchmarkNUMACopy(int src_node, int dst_node, Opts &opts) {
  * @param opts A reference to an Opts object containing various options and configurations for the benchmark.
  */
 double RunCPUCopyBenchmark(int src_node, int dst_node, Opts &opts) {
-    double max_time_ns = 0;
-
     // Run warm up rounds
     for (int i = 0; i < opts.num_warm_up; i++) {
         BenchmarkNUMACopy(src_node, dst_node, opts);
     }
 
+    double time_used_ns = 0;
+
     for (int i = 0; i < opts.num_loops; i++) {
-        double time_used_ns = BenchmarkNUMACopy(src_node, dst_node, opts);
-        max_time_ns = std::max(max_time_ns, time_used_ns);
+        time_used_ns += BenchmarkNUMACopy(src_node, dst_node, opts);
     }
 
-    return max_time_ns;
+    return time_used_ns / opts.num_loops;
 }
 
 int main(int argc, char **argv) {
diff --git a/tests/benchmarks/micro_benchmarks/test_cpu_memory_bw_latency_performance.py b/tests/benchmarks/micro_benchmarks/test_cpu_memory_bw_latency_performance.py
index 0a2d2a21f..df3ac6347 100644
--- a/tests/benchmarks/micro_benchmarks/test_cpu_memory_bw_latency_performance.py
+++ b/tests/benchmarks/micro_benchmarks/test_cpu_memory_bw_latency_performance.py
@@ -3,6 +3,8 @@
 
 """Tests for cpu-memory-bw-latency benchmark."""
 
+import os
+import platform
 import unittest
 
 from tests.helper.testcase import BenchmarkTestCase
@@ -148,3 +150,43 @@ def test_cpu_mem_bw_latency_benchmark_result_parsing(self):
         # Negative case - invalid raw output.
         assert (benchmark._process_raw_result(0, 'Invalid raw output') is False)
         assert (benchmark.return_code == ReturnCode.MICROBENCHMARK_RESULT_PARSING_FAILURE)
+
+    def test_preprocess_mlc(self):
+        """Test _preprocess method for Intel MLC tool."""
+        benchmark_name = 'cpu-memory-bw-latency'
+        (benchmark_class, predefine_params) = BenchmarkRegistry._BenchmarkRegistry__select_benchmark(benchmark_name, Platform.CPU)
+        assert (benchmark_class)
+
+        benchmark = benchmark_class(benchmark_name, parameters='--tests bandwidth_matrix')
+        benchmark._args = benchmark._parser.parse_args(['--bin_dir', '/mock/bin/dir'])
+        benchmark._bin_name = 'mlc'
+        benchmark._commands = []
+
+        # Mock os.access to return True
+        os.access = lambda path, mode: True
+
+        ret = benchmark._preprocess()
+        assert (ret is True)
+        assert (benchmark.return_code == ReturnCode.SUCCESS)
+        assert (len(benchmark._commands) == 1)
+        assert ('mlc --bandwidth_matrix;' in benchmark._commands[0])
+
+    def test_preprocess_general(self):
+        """Test _preprocess method for general CPU copy benchmark."""
+        benchmark_name = 'cpu-memory-bw-latency'
+        (benchmark_class, predefine_params) = BenchmarkRegistry._BenchmarkRegistry__select_benchmark(benchmark_name, Platform.CPU)
+        assert (benchmark_class)
+
+        benchmark = benchmark_class(benchmark_name, parameters='--size 1024 --num_warm_up 10 --num_loops 50 --check_data')
+        benchmark._args = benchmark._parser.parse_args(['--bin_dir', '/mock/bin/dir'])
+        benchmark._bin_name = 'cpu_copy'
+        benchmark._commands = []
+
+        # Mock platform.machine to return non-x86_64
+        platform.machine = lambda: 'arm64'
+
+        ret = benchmark._preprocess()
+        assert (ret is True)
+        assert (benchmark.return_code == ReturnCode.SUCCESS)
+        assert (len(benchmark._commands) == 1)
+        assert ('cpu_copy --size 1024 --num_warm_up 10 --num_loops 50 --check_data' in benchmark._commands[0])