diff --git a/cocos/2d/CCFastTMXLayer.cpp b/cocos/2d/CCFastTMXLayer.cpp
index 4f7ef7956d..169df9c211 100644
--- a/cocos/2d/CCFastTMXLayer.cpp
+++ b/cocos/2d/CCFastTMXLayer.cpp
@@ -128,6 +128,11 @@ FastTMXLayer::~FastTMXLayer()
     CC_SAFE_DELETE_ARRAY(_tiles);
 }
 
+bool sortQuadCommand(const V3F_C4B_T2F_Quad& a, const V3F_C4B_T2F_Quad& b)
+{
+    return a.bl.vertices.z < b.bl.vertices.z;
+}
+
 void FastTMXLayer::draw(Renderer *renderer, const Mat4& transform, uint32_t flags)
 {
     if( flags != 0 || _dirty )
@@ -140,7 +145,7 @@ void FastTMXLayer::draw(Renderer *renderer, const Mat4& transform, uint32_t flag
         rect = RectApplyTransform(rect, inv);
         
         _verticesToDraw = updateTiles(rect, renderer);
-        
+        std::sort(_quads.begin(), _quads.end(), sortQuadCommand);
         // don't draw more than 65535 vertices since we are using GL_UNSIGNED_SHORT for indices
         _verticesToDraw = std::min(_verticesToDraw, 65535);
         
@@ -148,13 +153,22 @@ void FastTMXLayer::draw(Renderer *renderer, const Mat4& transform, uint32_t flag
         
     }
     
-    auto glprogramState = GLProgramState::getOrCreateWithGLProgramName(GLProgram::SHADER_NAME_POSITION_TEXTURE_COLOR_NO_MVP);
-    for( int index = 0; index < _verticesToDraw/6; ++index)
+    auto glprogramState = GLProgramState::getOrCreateWithGLProgramName(GLProgram::SHADER_NAME_POSITION_TEXTURE_COLOR);
+    int index = 0;
+    for(const auto& quadNumberIter : _quadsNumber)
     {
-        float z = _quads[index].bl.vertices.z;
-        _renderCommands[index].init(z, _texture->getName(), glprogramState, BlendFunc::ALPHA_PREMULTIPLIED, &_quads[index], 1, transform);
-        renderer->addCommand(&_renderCommands[index]);
+        int z = quadNumberIter.first;
+        auto& quadCommand = _renderCommands2[quadNumberIter.first];
+        quadCommand.init(z, _texture->getName(), glprogramState, BlendFunc::ALPHA_PREMULTIPLIED, &_quads[index], quadNumberIter.second, transform);
+        index += quadNumberIter.second;
+        renderer->addCommand(&quadCommand);
     }
+//    for( int index = 0; index < _verticesToDraw/6; ++index)
+//    {
+//        float z = _quads[index].bl.vertices.z;
+//        _renderCommands[index].init(z, _texture->getName(), glprogramState, BlendFunc::ALPHA_PREMULTIPLIED, &_quads[index], 1, transform);
+//        renderer->addCommand(&_renderCommands[index]);
+//    }
     
 //    _customCommand.init(_globalZOrder);
 //    _customCommand.func = CC_CALLBACK_0(FastTMXLayer::onDraw, this, transform, flags);
@@ -218,6 +232,9 @@ int FastTMXLayer::updateTiles(const Rect& culledRect, Renderer* renderer)
         _indices.resize(quadsNeed * 6);
     }
 
+    //clear quadsNumber
+    _quadsNumber.clear();
+
     Size texSize = _tileSet->_imageSize;
     for (int y =  visibleTiles.origin.y - tilesOverY; y < visibleTiles.origin.y + visibleTiles.size.height + tilesOverY; ++y)
     {
@@ -242,7 +259,18 @@ int FastTMXLayer::updateTiles(const Rect& culledRect, Renderer* renderer)
                 float left, right, top, bottom, z;
                 
                 z = getVertexZForPos(Vec2(x, y));
-
+                
+                //add quadsNumber
+                auto quadsNumberIter = _quadsNumber.find(z);
+                if(quadsNumberIter == _quadsNumber.end())
+                {
+                    _quadsNumber.insert(std::make_pair(z, 1));
+                }
+                else
+                {
+                    quadsNumberIter->second += 1;
+                }
+                
                 // vertices
                 if (tileGID & kTMXTileDiagonalFlag)
                 {
diff --git a/cocos/2d/CCFastTMXLayer.h b/cocos/2d/CCFastTMXLayer.h
index 052c6a8184..7291921f4c 100644
--- a/cocos/2d/CCFastTMXLayer.h
+++ b/cocos/2d/CCFastTMXLayer.h
@@ -34,6 +34,7 @@ THE SOFTWARE.
 #include "renderer/CCQuadCommand.h"
 
 #include <map>
+#include <unordered_map>
 
 NS_CC_BEGIN
 
@@ -245,6 +246,9 @@ protected:
     /** quads to be rendered */
     std::vector<V3F_C4B_T2F_Quad> _quads;
     std::vector<QuadCommand> _renderCommands;
+    
+    std::map<int/*zorder*/, ssize_t/*number*/> _quadsNumber;
+    std::unordered_map<int/*zorder*/, NopreMultiplyMVQuadCommand> _renderCommands2;
     /** indices */
     std::vector<GLushort> _indices;
     bool _dirty;
diff --git a/cocos/base/CCDirector.cpp b/cocos/base/CCDirector.cpp
index 9d1e02f187..a2d22b46e9 100644
--- a/cocos/base/CCDirector.cpp
+++ b/cocos/base/CCDirector.cpp
@@ -1066,7 +1066,7 @@ void Director::showStats()
     {
         char buffer[30];
 
-        if (_accumDt > CC_DIRECTOR_STATS_INTERVAL)
+        if (_accumDt > CC_DIRECTOR_STATS_INTERVAL * 20)
         {
             _frameRate = _frames / _accumDt;
             _frames = 0;
diff --git a/cocos/renderer/CCQuadCommand.cpp b/cocos/renderer/CCQuadCommand.cpp
index 4b497ff253..eb4fe6e66e 100644
--- a/cocos/renderer/CCQuadCommand.cpp
+++ b/cocos/renderer/CCQuadCommand.cpp
@@ -40,6 +40,7 @@ QuadCommand::QuadCommand()
 ,_blendType(BlendFunc::DISABLE)
 ,_quads(nullptr)
 ,_quadsCount(0)
+,_preMultiplyMV(true)
 {
     _type = RenderCommand::Type::QUAD_COMMAND;
 }
diff --git a/cocos/renderer/CCQuadCommand.h b/cocos/renderer/CCQuadCommand.h
index 8f3d3622f1..7024a2050f 100644
--- a/cocos/renderer/CCQuadCommand.h
+++ b/cocos/renderer/CCQuadCommand.h
@@ -38,7 +38,7 @@ public:
     static const int MATERIAL_ID_DO_NOT_BATCH = 0;
 
     QuadCommand();
-    ~QuadCommand();
+    virtual ~QuadCommand();
 
     /** Initializes the command with a globalZOrder, a texture ID, a `GLProgram`, a blending function, a pointer to quads,
      * quantity of quads, and the Model View transform to be used for the quads */
@@ -55,10 +55,10 @@ public:
     inline GLProgramState* getGLProgramState() const { return _glProgramState; }
     inline BlendFunc getBlendType() const { return _blendType; }
     inline const Mat4& getModelView() const { return _mv; }
-    
+    inline bool getPremultiplyMV() const { return _preMultiplyMV; }
 
 protected:
-    void generateMaterialID();
+    virtual void generateMaterialID();
 
     uint32_t _materialID;
     GLuint _textureID;
@@ -67,7 +67,29 @@ protected:
     V3F_C4B_T2F_Quad* _quads;
     ssize_t _quadsCount;
     Mat4 _mv;
+    bool _preMultiplyMV;
 };
+
+class NopreMultiplyMVQuadCommand : public QuadCommand
+{
+public:
+    NopreMultiplyMVQuadCommand()
+    :QuadCommand()
+    {
+        _preMultiplyMV = false;
+    }
+    
+    virtual ~NopreMultiplyMVQuadCommand()
+    {
+    }
+    
+protected:
+    virtual void generateMaterialID()
+    {
+        _materialID = QuadCommand::MATERIAL_ID_DO_NOT_BATCH;
+    }
+};
+
 NS_CC_END
 
 #endif //_CC_QUADCOMMAND_H_
diff --git a/cocos/renderer/CCRenderer.cpp b/cocos/renderer/CCRenderer.cpp
index 222118da6f..5a5964a28d 100644
--- a/cocos/renderer/CCRenderer.cpp
+++ b/cocos/renderer/CCRenderer.cpp
@@ -296,7 +296,13 @@ void Renderer::visitRenderQueue(const RenderQueue& queue)
             _batchedQuadCommands.push_back(cmd);
             
             memcpy(_quads + _numQuads, cmd->getQuads(), sizeof(V3F_C4B_T2F_Quad) * cmd->getQuadCount());
-            convertToWorldCoordinates(_quads + _numQuads, cmd->getQuadCount(), cmd->getModelView());
+            
+            //only multiply mv when getPremultiplyMV is true
+            if(cmd->getPremultiplyMV())
+            {
+                convertToWorldCoordinates(_quads + _numQuads, cmd->getQuadCount(), cmd->getModelView());
+            }
+            
             
             _numQuads += cmd->getQuadCount();