diff --git a/include/AMMBench.h b/include/AMMBench.h
index 84e5da35..757fffea 100755
--- a/include/AMMBench.h
+++ b/include/AMMBench.h
@@ -99,6 +99,7 @@
 #include <CPPAlgos/AbstractCPPAlgo.h>
 #include <CPPAlgos/CPPAlgoTable.h>
 #include <CPPAlgos/CRSCPPAlgo.h>
+#include <CPPAlgos/CRSV2CPPAlgo.h>
 /**
  * @}
  *
diff --git a/include/CPPAlgos/CRSV2CPPAlgo.h b/include/CPPAlgos/CRSV2CPPAlgo.h
new file mode 100644
index 00000000..c83fedc5
--- /dev/null
+++ b/include/CPPAlgos/CRSV2CPPAlgo.h
@@ -0,0 +1,58 @@
+//
+// Created by haolan on 5/26/23.
+//
+
+#ifndef INTELLISTREAM_CRSV2CPPALGO_H
+#define INTELLISTREAM_CRSV2CPPALGO_H
+
+#include <CPPAlgos/AbstractCPPAlgo.h>
+
+namespace AMMBench {
+/**
+ * @ingroup AMMBENCH_CppAlgos The algorithms writtrn in c++
+ * @{
+ */
+/**
+ * @class CRSCPPlgo CPPAlgos/CRSV2CPPAlgo.h
+ * @brief The cloumn row sampling (CRS) class of c++ algos
+ *
+ */
+    class CRSV2CPPAlgo : public AMMBench::AbstractCPPAlgo {
+    public:
+        CRSV2CPPAlgo() {
+
+        }
+
+        ~CRSV2CPPAlgo() {
+
+        }
+
+        /**
+         * @brief the virtual function provided for outside callers, rewrite in children classes
+         * @param A the A matrix
+         * @param B the B matrix
+         * @param sketchSize the size of sketc or sampling
+         * @return the output c matrix
+         */
+        virtual torch::Tensor amm(torch::Tensor A, torch::Tensor B, int sketchSize);
+
+    };
+
+/**
+ * @ingroup AMMBENCH_CppAlgos
+ * @typedef AbstractMatrixCppAlgoPtr
+ * @brief The class to describe a shared pointer to @ref CRSV2CppAlgo
+
+ */
+    typedef std::shared_ptr<class AMMBench::CRSV2CPPAlgo> CRSV2CPPAlgoPtr;
+/**
+ * @ingroup AMMBENCH_CppAlgos
+ * @def newCRSV2CppAlgo
+ * @brief (Macro) To creat a new @ref  CRSV2CppAlgounder shared pointer.
+ */
+#define newCRSV2CPPAlgo std::make_shared<AMMBench::CRSV2CPPAlgo>
+}
+/**
+ * @}
+ */
+#endif //INTELLISTREAM_CRSV2CPPALGO_H
diff --git a/src/CPPAlgos/CMakeLists.txt b/src/CPPAlgos/CMakeLists.txt
index f8805edd..79f5a9d4 100644
--- a/src/CPPAlgos/CMakeLists.txt
+++ b/src/CPPAlgos/CMakeLists.txt
@@ -2,4 +2,5 @@ add_sources(
         AbstractCPPAlgo.cpp
         CRSCPPAlgo.cpp
         CPPAlgoTable.cpp
+        CRSV2CPPAlgo.cpp
 )
\ No newline at end of file
diff --git a/src/CPPAlgos/CPPAlgoTable.cpp b/src/CPPAlgos/CPPAlgoTable.cpp
index 572856a0..af137843 100644
--- a/src/CPPAlgos/CPPAlgoTable.cpp
+++ b/src/CPPAlgos/CPPAlgoTable.cpp
@@ -4,10 +4,12 @@
 
 #include <CPPAlgos/CPPAlgoTable.h>
 #include <CPPAlgos/CRSCPPAlgo.h>
+#include <CPPAlgos/CRSV2CPPAlgo.h>
 namespace AMMBench {
 AMMBench::CPPAlgoTable::CPPAlgoTable() {
   algoMap["mm"] = newAbstractCPPAlgo();
   algoMap["crs"] = newCRSCPPAlgo();
+    algoMap["crsV2"] = newCRSV2CPPAlgo();
 }
 
 } // AMMBench
\ No newline at end of file
diff --git a/src/CPPAlgos/CRSV2CPPAlgo.cpp b/src/CPPAlgos/CRSV2CPPAlgo.cpp
new file mode 100644
index 00000000..01378f76
--- /dev/null
+++ b/src/CPPAlgos/CRSV2CPPAlgo.cpp
@@ -0,0 +1,40 @@
+//
+// Created by haolan on 5/26/23.
+//
+
+#include <CPPAlgos/CRSV2CPPAlgo.h>
+
+namespace AMMBench {
+    torch::Tensor AMMBench::CRSV2CPPAlgo::amm(torch::Tensor A, torch::Tensor B, int k) {
+        A = A.t();
+        auto A_size = A.sizes();
+        int64_t n = A_size[0];
+        // int64_t m = A_size[1];
+
+        assert(n == B.size(0));
+        //TORCH_CHECK(n == B.size(0));
+        //TORCH_CHECK(k < n);
+
+        //INTELLI_INFO("Running CRS V2 CPP");
+
+        // probability distribution
+        torch::Tensor sample = torch::rand({n}); // default: uniform
+
+        // diagonal scaling matrix D (nxn)
+        sample = sample.div(sample.sum());
+        torch::Tensor D = torch::diag(1.0 / torch::sqrt(k * sample));
+
+        // sampling matrix S (kxn)
+        torch::Tensor column_indices = torch::multinomial(sample, k, true);
+        torch::Tensor S = torch::zeros({k, n});
+        for (int64_t row = 0; row < k; row++) {
+            int64_t col = column_indices[row].item<int64_t>();
+            S[row][col] = 1;
+        }
+
+        torch::Tensor a = torch::matmul(torch::matmul(A.t(), D), S.t());
+        torch::Tensor b = torch::matmul(torch::matmul(a, S), D);
+
+        return torch::matmul(b, B);
+    }
+} // AMMBench
\ No newline at end of file
diff --git a/test/SystemTest/CRSTest.cpp b/test/SystemTest/CRSTest.cpp
index 168cdd85..33c4add2 100644
--- a/test/SystemTest/CRSTest.cpp
+++ b/test/SystemTest/CRSTest.cpp
@@ -61,7 +61,7 @@ TEST_CASE("Test the COLUMN ROW SAMPLINGS, V2", "[short]")
   // place your test here
   REQUIRE(a == 0);
 }
-TEST_CASE("Test pure cpp versiom", "[short]")
+TEST_CASE("Test CRS in cpp", "[short]")
 {
   torch::manual_seed(114514);
   AMMBench::CRSCPPAlgo crs;
@@ -71,4 +71,15 @@ TEST_CASE("Test pure cpp versiom", "[short]")
   auto ammC = crs.amm(A, B, 20);
   double froError = INTELLI::UtilityFunctions::relativeFrobeniusNorm(realC, ammC);
   REQUIRE(froError < 0.5);
+}
+TEST_CASE("Test CRS v2 in cpp", "[short]")
+{
+    torch::manual_seed(114514);
+    AMMBench::CRSV2CPPAlgo crs;
+    auto A = torch::rand({400, 400});
+    auto B = torch::rand({400, 400});
+    auto realC = torch::matmul(A, B);
+    auto ammC = crs.amm(A, B, 20);
+    double froError = INTELLI::UtilityFunctions::relativeFrobeniusNorm(realC, ammC);
+    REQUIRE(froError < 0.5);
 }
\ No newline at end of file