diff --git a/src/lv_core/lv_obj.c b/src/lv_core/lv_obj.c
index fa0ad53ac..dbc630928 100644
--- a/src/lv_core/lv_obj.c
+++ b/src/lv_core/lv_obj.c
@@ -38,6 +38,10 @@
 
 #include LV_THEME_DEFAULT_INCLUDE
 
+#if LV_USE_GPU_STM32_DMA2D
+#include "../lv_gpu/lv_gpu_stm32_dma2d.h"
+#endif
+
 /*********************
  *      DEFINES
  *********************/
@@ -139,6 +143,11 @@ void lv_init(void)
     _lv_group_init();
 #endif
 
+#if LV_USE_GPU_STM32_DMA2D
+    /*Initialize DMA2D GPU*/
+    lv_gpu_stm32_dma2d_init();
+#endif
+
     _lv_ll_init(&LV_GC_ROOT(_lv_obj_style_trans_ll), sizeof(lv_style_trans_t));
     
     _lv_ll_init(&LV_GC_ROOT(_lv_disp_ll), sizeof(lv_disp_t));
diff --git a/src/lv_draw/lv_draw_blend.c b/src/lv_draw/lv_draw_blend.c
index ea446c765..3bdd7e0d7 100644
--- a/src/lv_draw/lv_draw_blend.c
+++ b/src/lv_draw/lv_draw_blend.c
@@ -338,7 +338,7 @@ LV_ATTRIBUTE_FAST_MEM static void fill_normal(const lv_area_t * disp_area, lv_co
         else {
 #if LV_USE_GPU
             if(disp->driver.gpu_blend_cb && lv_area_get_size(draw_area) > GPU_SIZE_LIMIT) {
-                LV_BLEND_BUF_ATTR static lv_color_t blend_buf[LV_HOR_RES_MAX];
+                LV_DMA_ATTR static lv_color_t blend_buf[LV_HOR_RES_MAX];
                 for(x = 0; x < draw_area_w ; x++) blend_buf[x].full = color.full;
 
                 for(y = draw_area->y1; y <= draw_area->y2; y++) {
@@ -353,7 +353,7 @@ LV_ATTRIBUTE_FAST_MEM static void fill_normal(const lv_area_t * disp_area, lv_co
 #if LV_USE_GPU_STM32_DMA2D
             if(lv_area_get_size(draw_area) >= 240) {
                 /* blend_buf MUST be in DMA accessible RAM (depending on linker this isn't always the case - CCMRAM) */
-                LV_BLEND_BUF_ATTR static lv_color_t blend_buf[LV_HOR_RES_MAX] = {0};
+                LV_DMA_ATTR static lv_color_t blend_buf[LV_HOR_RES_MAX] = {0};
                 if(blend_buf[0].full != color.full) lv_color_fill(blend_buf, color, LV_HOR_RES_MAX);
 
                 lv_coord_t line_h = LV_HOR_RES_MAX / draw_area_w;
diff --git a/src/lv_gpu/lv_gpu_stm32_dma2d.c b/src/lv_gpu/lv_gpu_stm32_dma2d.c
index a756e3af5..7e7367331 100644
--- a/src/lv_gpu/lv_gpu_stm32_dma2d.c
+++ b/src/lv_gpu/lv_gpu_stm32_dma2d.c
@@ -11,7 +11,7 @@
 
 #if LV_USE_GPU_STM32_DMA2D
 
-#include "stm32f429xx.h"
+#include LV_CPU_DMA2D_CMSIS_INCLUDE
 
 /*********************
  *      DEFINES
@@ -64,6 +64,9 @@ void lv_gpu_stm32_dma2d_init(void)
     /* Enable DMA2D clock */
     RCC->AHB1ENR |= RCC_AHB1ENR_DMA2DEN;
 
+    /* Delay after setting peripheral clock */
+    volatile uint32_t temp = RCC->AHB1ENR;
+
     /* set output colour mode */
     DMA2D->OPFCCR = DMA2D_COLOR_FORMAT;
 }
diff --git a/src/lv_gpu/lv_gpu_stm32_dma2d.h b/src/lv_gpu/lv_gpu_stm32_dma2d.h
index 52e4b2fe4..035647199 100644
--- a/src/lv_gpu/lv_gpu_stm32_dma2d.h
+++ b/src/lv_gpu/lv_gpu_stm32_dma2d.h
@@ -34,6 +34,11 @@ extern "C" {
  * GLOBAL PROTOTYPES
  **********************/
 
+/**
+ * Turn on the peripheral and set output color mode, this only needs to be done once
+ */
+void lv_gpu_stm32_dma2d_init(void);
+
 /**
  * Fill an area in the buffer with a color
  * @param buf a buffer which should be filled