diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index c98cbcbbebd0c..bebb7f3db8bad 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -157,15 +157,15 @@ jobs:
       matrix:
         include:
           - build: 'avx2'
-            defines: '-DLLAMA_BUILD_SERVER=ON'
+            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_NATIVE=OFF'
           - build: 'avx'
-            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX2=OFF'
+            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_NATIVE=OFF -DLLAMA_AVX2=OFF'
           - build: 'avx512'
-            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX512=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_NATIVE=OFF -DLLAMA_AVX512=ON -DBUILD_SHARED_LIBS=ON'
           - build: 'clblast'
-            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_CLBLAST=ON -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/clblast"'
+            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_NATIVE=OFF -DLLAMA_CLBLAST=ON -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/clblast"'
           - build: 'openblas'
-            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
+            defines: '-DLLAMA_BUILD_SERVER=ON -DLLAMA_NATIVE=OFF -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
 
     steps:
       - name: Clone
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 21f4ec9ddd267..531240b02af5a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -38,7 +38,7 @@ endif()
 
 # general
 option(LLAMA_STATIC                     "llama: static link libraries"                          OFF)
-option(LLAMA_NATIVE                     "llama: enable -march=native flag"                      OFF)
+option(LLAMA_NATIVE                     "llama: enable -march=native flag"                      ON)
 option(LLAMA_LTO                        "llama: enable link time optimization"                  OFF)
 
 # debug
@@ -109,6 +109,10 @@ else()
     message(WARNING "Git repository not found; to enable automatic generation of build info, make sure Git is installed and the project is a Git repository.")
 endif()
 
+IF(LLAMA_NATIVE)
+    include(cmake/FindSIMD.cmake)
+ENDIF()
+
 #
 # Compile flags
 #
diff --git a/cmake/FindSIMD.cmake b/cmake/FindSIMD.cmake
new file mode 100644
index 0000000000000..fc8b079bc3167
--- /dev/null
+++ b/cmake/FindSIMD.cmake
@@ -0,0 +1,99 @@
+INCLUDE(CheckCSourceRuns)
+
+SET(AVX_CODE "
+  #include <immintrin.h>
+  int main()
+  {
+    __m256 a;
+    a = _mm256_set1_ps(0);
+    return 0;
+  }
+")
+
+SET(AVX512_CODE "
+  #include <immintrin.h>
+  int main()
+  {
+    __m512i a = _mm512_set_epi8(0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0);
+    __m512i b = a;
+    __mmask64 equality_mask = _mm512_cmp_epi8_mask(a, b, _MM_CMPINT_EQ);
+    return 0;
+  }
+")
+
+SET(AVX2_CODE "
+  #include <immintrin.h>
+  int main()
+  {
+    __m256i a = {0};
+    a = _mm256_abs_epi16(a);
+    __m256i x;
+    _mm256_extract_epi64(x, 0); // we rely on this in our AVX2 code
+    return 0;
+  }
+")
+
+SET(FMA_CODE "
+  #include <immintrin.h>
+  int main()
+  {
+    __m256 acc = _mm256_setzero_ps();
+    const __m256 d = _mm256_setzero_ps();
+    const __m256 p = _mm256_setzero_ps();
+    acc = _mm256_fmadd_ps( d, p, acc );
+    return 0;
+  }
+")
+
+MACRO(CHECK_SSE type flags)
+  SET(__FLAG_I 1)
+  SET(CMAKE_REQUIRED_FLAGS_SAVE ${CMAKE_REQUIRED_FLAGS})
+  FOREACH(__FLAG ${flags})
+    IF(NOT ${type}_FOUND)
+      SET(CMAKE_REQUIRED_FLAGS ${__FLAG})
+      CHECK_C_SOURCE_RUNS("${${type}_CODE}" HAS_${type}_${__FLAG_I})
+      IF(HAS_${type}_${__FLAG_I})
+        SET(${type}_FOUND TRUE CACHE BOOL "${type} support")
+        SET(${type}_FLAGS "${__FLAG}" CACHE STRING "${type} flags")
+      ENDIF()
+      MATH(EXPR __FLAG_I "${__FLAG_I}+1")
+    ENDIF()
+  ENDFOREACH()
+  SET(CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS_SAVE})
+
+  IF(NOT ${type}_FOUND)
+    SET(${type}_FOUND FALSE CACHE BOOL "${type} support")
+    SET(${type}_FLAGS "" CACHE STRING "${type} flags")
+  ENDIF()
+
+  MARK_AS_ADVANCED(${type}_FOUND ${type}_FLAGS)
+
+ENDMACRO()
+
+CHECK_SSE("AVX" " ;/arch:AVX")
+IF(NOT ${AVX_FOUND})
+    set(LLAMA_AVX OFF)
+ELSE()
+    set(LLAMA_AVX ON)
+ENDIF()
+
+CHECK_SSE("AVX2" " ;/arch:AVX2")
+IF(NOT ${AVX2_FOUND})
+    set(LLAMA_AVX2 OFF)
+ELSE()
+    set(LLAMA_AVX2 ON)
+ENDIF()
+
+CHECK_SSE("AVX512" " ;/arch:AVX512")
+IF(NOT ${AVX512_FOUND})
+    set(LLAMA_AVX512 OFF)
+ELSE()
+    set(LLAMA_AVX512 ON)
+ENDIF()