diff --git a/tests/benchmark/parser/CMakeLists.txt b/tests/benchmark/parser/CMakeLists.txt
deleted file mode 100644
index 47d1219321ff..000000000000
--- a/tests/benchmark/parser/CMakeLists.txt
+++ /dev/null
@@ -1,18 +0,0 @@
-cmake_minimum_required(VERSION 3.0)
-
-project(benchmark)
-
-OPTION(USE_PRECISE_TEXT_PARSER "Use precise (and faster) double parser for text input file" OFF)
-
-if(USE_PRECISE_TEXT_PARSER)
-    ADD_DEFINITIONS(-DUSE_PRECISE_TEXT_PARSER)
-endif(USE_PRECISE_TEXT_PARSER)
-
-set(CMAKE_CXX_STANDARD 11)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -O3")
-
-include_directories(${CMAKE_SOURCE_DIR}/../../../include)
-include_directories(${CMAKE_SOURCE_DIR}/../../../src)
-#link_directories(${CMAKE_SOURCE_DIR}/lib)
-
-add_executable(parser parser.cpp)
diff --git a/tests/benchmark/parser/README.md b/tests/benchmark/parser/README.md
deleted file mode 100644
index c5c3f1edfdf4..000000000000
--- a/tests/benchmark/parser/README.md
+++ /dev/null
@@ -1,38 +0,0 @@
-This is a simple benchmark comparing performance of `Common::Atof`
-and `Common::AtofPrecise` when used in `CSVParser`.
-
-Just run `./run_parser_benchmark.sh` in this directory.
-
-The test script generates 20000 rows, 2000 columns csv, 840MB file size.
-
-For this test, `Common::Atof` is much faster than `Common::AtofPrecise`.
-
-Benchmark run output on Intel Xeon 2640 v3:
-
-```
-========== Benchmark run Atof parser ==========                                                                                                                                                                                                                                                                                                                                                                 real    0m2.027s                                                                                                                                                                                        user    0m1.822s
-real    0m2.027s
-user    0m1.822s
-sys     0m0.204s
-
-real    0m2.186s
-user    0m1.998s
-sys     0m0.188s
-
-real    0m2.202s
-user    0m2.010s
-sys     0m0.192s
-
-========== Benchmark run AtofPrecise parser ==========
-real    0m6.556s
-user    0m6.324s
-sys     0m0.232s
-
-real    0m6.648s
-user    0m6.496s
-sys     0m0.152s
-
-real    0m6.912s
-user    0m6.748s
-sys     0m0.164s
-```
\ No newline at end of file
diff --git a/tests/benchmark/parser/gen_csv.py b/tests/benchmark/parser/gen_csv.py
deleted file mode 100644
index 5c2c9b5d2e23..000000000000
--- a/tests/benchmark/parser/gen_csv.py
+++ /dev/null
@@ -1,17 +0,0 @@
-import sys
-
-import numpy as np
-
-
-def gen_csv(fname, nrow, ncol):
-    nrow = int(nrow)
-    ncol = int(ncol)
-
-    arr = np.random.random(nrow * ncol) * 5
-    arr = arr.reshape((nrow, ncol))
-    np.savetxt(fname, arr, fmt='%.19f', delimiter=',')
-
-
-if __name__ == '__main__':
-    import argh
-    argh.dispatch_command(gen_csv)
\ No newline at end of file
diff --git a/tests/benchmark/parser/parser.cpp b/tests/benchmark/parser/parser.cpp
deleted file mode 100644
index 6eb244ef707c..000000000000
--- a/tests/benchmark/parser/parser.cpp
+++ /dev/null
@@ -1,49 +0,0 @@
-// This is a very simple benchmark for comparing performance of Atof and AtofPrecise.
-
-#include <cstdlib>
-#include <string>
-#include <fstream>
-#include <iostream>
-
-#include <io/parser.hpp>
-
-namespace LightGBM {
-
-void ParseCSV(const std::string& fpath, int ncol) {
-  CSVParser parser(-1, ncol);
-
-  std::ifstream infile(fpath);
-  if (! infile) {
-    std::cerr << "fail to open " << fpath;
-    std::exit(1);
-  }
-
-  std::string line;
-  double label;
-  std::vector<std::pair<int, double>> oneline_features;
-  while (getline(infile, line)) {
-    parser.ParseOneLine(line.c_str(), &oneline_features, &label);
-//    printf("%f\n", oneline_features[0].second);
-    oneline_features.clear();
-  }
-}
-
-}  // namespace LightGBM
-
-int main(int argc, const char* argv[]) {
-  if (argc != 3) {
-    printf("usage: parser <fname> <ncol>\n");
-    exit(1);
-  }
-
-  const char* fpath = argv[1];
-  long ncol = strtol(argv[2], nullptr, 10);
-  if (errno != 0) {
-    fprintf(stderr, "fail to parse ncol\n");
-    exit(1);
-  }
-
-  LightGBM::ParseCSV(fpath, ncol);
-
-  return 0;
-}
\ No newline at end of file
diff --git a/tests/benchmark/parser/run_parser_benchmark.sh b/tests/benchmark/parser/run_parser_benchmark.sh
deleted file mode 100755
index 2d4d380fac71..000000000000
--- a/tests/benchmark/parser/run_parser_benchmark.sh
+++ /dev/null
@@ -1,43 +0,0 @@
-#!/bin/bash
-
-set -e
-
-nrow=20000
-ncol=2000
-
-build() {
-  d=$1
-  flag=$2
-  echo "building $d $flag"
-  test -d $d || (mkdir -p $d && cd $d && cmake $flag ..)
-  pushd $d
-  make
-  popd
-}
-
-gen_data() {
-  if [[ ! -f test.csv ]]; then
-    echo "generating csv ..."
-    python gen_csv.py test.csv $nrow $ncol
-  fi
-}
-
-parser_benchmark() {
-  echo "========== Benchmark run Atof parser =========="
-  for i in {1..3}; do
-#    /usr/bin/time ./build/parser test.csv $ncol
-    time ./build/parser test.csv $ncol
-  done
-
-  echo
-  echo "========== Benchmark run AtofPrecise parser =========="
-  for i in {1..3}; do
-#    /usr/bin/time ./build-precise/parser test.csv $ncol
-    time ./build-precise/parser test.csv $ncol
-  done
-}
-
-build build ""
-build build-precise "-DUSE_PRECISE_TEXT_PARSER=on"
-gen_data
-parser_benchmark