diff --git a/src/cnn/train.c b/src/cnn/train.c
index ad971dd..72cd5a7 100644
--- a/src/cnn/train.c
+++ b/src/cnn/train.c
@@ -127,7 +127,7 @@ void* train_thread(void* parameters) {
 
 void train(int dataset_type, char* images_file, char* labels_file, char* data_dir, int epochs, char* out, char* recover) {
     #ifdef USE_CUDA
-    bool compatibility = check_cuda_compatibility();
+    bool compatibility = cuda_setup(true);
     if (!compatibility) {
         printf("Exiting.\n");
         exit(1);
diff --git a/src/common/include/utils.h b/src/common/include/utils.h
index 0e4603f..8c66828 100644
--- a/src/common/include/utils.h
+++ b/src/common/include/utils.h
@@ -60,8 +60,10 @@ extern "C"
 #endif
 /*
 * Vérification de la compatibilité CUDA
+* spécifier avec "verbose" si il faut afficher
+* la carte utilisée notamment
 */
-bool check_cuda_compatibility();
+bool cuda_setup(bool verbose);
 
 #ifdef __CUDACC__
 extern "C"
diff --git a/src/common/utils.c b/src/common/utils.c
index 93b1671..5292dc3 100644
--- a/src/common/utils.c
+++ b/src/common/utils.c
@@ -1,6 +1,7 @@
 #include <stdbool.h>
 #include <stdlib.h>
 #include <stdio.h>
+
 #ifdef USE_CUDA
     #ifndef __CUDACC__
         #include "cuda_runtime.h"
@@ -42,39 +43,65 @@ int i_div_up(int a, int b) { // Partie entière supérieure de a/b
 #ifdef __CUDACC__
 extern "C"
 #endif
-bool check_cuda_compatibility() {
+bool cuda_setup(bool verbose) {
     #ifdef __CUDACC__
     int nDevices;
+    int selected_device = 0;
+    cudaDeviceProp selected_prop;
     cudaDeviceProp prop;
 
     cudaGetDeviceCount(&nDevices);
-    if (nDevices == 0) {
-        printf("Pas d'utilisation du GPU\n\n");
+    if (nDevices <= 0) { // I've seen weird issues when there is no GPU at all
+        if (verbose) {
+            printf("Pas d'utilisation du GPU\n\n");
+        }
         return false;
     }
 
-    printf("GPUs disponibles:\n");
+    if (verbose) {
+        printf("GPUs disponibles:\n");
+    }
+
+    cudaGetDeviceProperties(&selected_prop, selected_device);
 
     for (int i=0; i < nDevices; i++) {
         cudaGetDeviceProperties(&prop, i);
-        printf(" - %s\n\t - Compute Capability: %d.%d\n\t - Memory available: ", prop.name, prop.major, prop.minor);
-        printf_memory(prop.totalGlobalMem);
-        printf("\n\t - Shared Memory per block: ");
-        printf_memory(prop.sharedMemPerBlock);
-        printf("\n\n");
+
+        if (verbose) {
+            printf(" - %s\n\t - Compute Capability: %d.%d\n\t - Memory available: ", prop.name, prop.major, prop.minor);
+            printf_memory(prop.totalGlobalMem);
+            printf("\n\t - Shared Memory per block: ");
+            printf_memory(prop.sharedMemPerBlock);
+            printf("\n\n");
+        }
+
+        if (prop.clockRate*prop.multiProcessorCount >= selected_prop.clockRate*selected_prop.multiProcessorCount) { // This criteria approximately matches the best device
+            selected_prop = prop;
+            selected_device = i;
+        }
     }
 
-    cudaGetDeviceProperties(&prop, 0);
-    printf("Utilisation du GPU: " BLUE "%s" RESET "\n\n", prop.name);
+    cudaSetDevice(selected_device); // Select the best device for computation
+    if (verbose) {
+        printf("Utilisation du GPU: " BLUE "%s" RESET "\n\n", selected_prop.name);
+    }
 
-    if (prop.sharedMemPerBlock != MEMORY_BLOCK) {
+    if (BLOCKSIZE_x*BLOCKSIZE_y*BLOCKSIZE_z > prop.maxThreadsPerBlock) {
+        printf_error((char*)"La taille de bloc sélectionnée est trop grande.\n");
+        printf("\tMaximum accepté: %d\n", selected_prop.maxThreadsPerBlock);
+        exit(1);
+    }
+    if (selected_prop.sharedMemPerBlock != MEMORY_BLOCK) { // C'est un warning, on l'affiche dans tous les cas
         printf_warning((char*)"La taille des blocs mémoire du GPU et celle utilisée dans le code diffèrent.\n");
         printf("\tCela peut mener à une utilisation supplémentaire de VRAM.\n");
-        printf("\tChanger MEMORY_BLOCK à %ld dans src/include/memory_management.h\n", prop.sharedMemPerBlock);
+        printf("\tChanger MEMORY_BLOCK à %ld dans src/include/memory_management.h\n", selected_prop.sharedMemPerBlock);
     }
     return true;
     #else
-    printf("Pas d'utilisation du GPU\n\n");
+    if (verbose) {
+        printf("Pas d'utilisation du GPU\n\n");
+    }
+
     return false;
     #endif
 }
diff --git a/src/common/utils.cu b/src/common/utils.cu
index 93b1671..5292dc3 100644
--- a/src/common/utils.cu
+++ b/src/common/utils.cu
@@ -1,6 +1,7 @@
 #include <stdbool.h>
 #include <stdlib.h>
 #include <stdio.h>
+
 #ifdef USE_CUDA
     #ifndef __CUDACC__
         #include "cuda_runtime.h"
@@ -42,39 +43,65 @@ int i_div_up(int a, int b) { // Partie entière supérieure de a/b
 #ifdef __CUDACC__
 extern "C"
 #endif
-bool check_cuda_compatibility() {
+bool cuda_setup(bool verbose) {
     #ifdef __CUDACC__
     int nDevices;
+    int selected_device = 0;
+    cudaDeviceProp selected_prop;
     cudaDeviceProp prop;
 
     cudaGetDeviceCount(&nDevices);
-    if (nDevices == 0) {
-        printf("Pas d'utilisation du GPU\n\n");
+    if (nDevices <= 0) { // I've seen weird issues when there is no GPU at all
+        if (verbose) {
+            printf("Pas d'utilisation du GPU\n\n");
+        }
         return false;
     }
 
-    printf("GPUs disponibles:\n");
+    if (verbose) {
+        printf("GPUs disponibles:\n");
+    }
+
+    cudaGetDeviceProperties(&selected_prop, selected_device);
 
     for (int i=0; i < nDevices; i++) {
         cudaGetDeviceProperties(&prop, i);
-        printf(" - %s\n\t - Compute Capability: %d.%d\n\t - Memory available: ", prop.name, prop.major, prop.minor);
-        printf_memory(prop.totalGlobalMem);
-        printf("\n\t - Shared Memory per block: ");
-        printf_memory(prop.sharedMemPerBlock);
-        printf("\n\n");
+
+        if (verbose) {
+            printf(" - %s\n\t - Compute Capability: %d.%d\n\t - Memory available: ", prop.name, prop.major, prop.minor);
+            printf_memory(prop.totalGlobalMem);
+            printf("\n\t - Shared Memory per block: ");
+            printf_memory(prop.sharedMemPerBlock);
+            printf("\n\n");
+        }
+
+        if (prop.clockRate*prop.multiProcessorCount >= selected_prop.clockRate*selected_prop.multiProcessorCount) { // This criteria approximately matches the best device
+            selected_prop = prop;
+            selected_device = i;
+        }
     }
 
-    cudaGetDeviceProperties(&prop, 0);
-    printf("Utilisation du GPU: " BLUE "%s" RESET "\n\n", prop.name);
+    cudaSetDevice(selected_device); // Select the best device for computation
+    if (verbose) {
+        printf("Utilisation du GPU: " BLUE "%s" RESET "\n\n", selected_prop.name);
+    }
 
-    if (prop.sharedMemPerBlock != MEMORY_BLOCK) {
+    if (BLOCKSIZE_x*BLOCKSIZE_y*BLOCKSIZE_z > prop.maxThreadsPerBlock) {
+        printf_error((char*)"La taille de bloc sélectionnée est trop grande.\n");
+        printf("\tMaximum accepté: %d\n", selected_prop.maxThreadsPerBlock);
+        exit(1);
+    }
+    if (selected_prop.sharedMemPerBlock != MEMORY_BLOCK) { // C'est un warning, on l'affiche dans tous les cas
         printf_warning((char*)"La taille des blocs mémoire du GPU et celle utilisée dans le code diffèrent.\n");
         printf("\tCela peut mener à une utilisation supplémentaire de VRAM.\n");
-        printf("\tChanger MEMORY_BLOCK à %ld dans src/include/memory_management.h\n", prop.sharedMemPerBlock);
+        printf("\tChanger MEMORY_BLOCK à %ld dans src/include/memory_management.h\n", selected_prop.sharedMemPerBlock);
     }
     return true;
     #else
-    printf("Pas d'utilisation du GPU\n\n");
+    if (verbose) {
+        printf("Pas d'utilisation du GPU\n\n");
+    }
+
     return false;
     #endif
 }
diff --git a/src/scripts/convolution_benchmark.cu b/src/scripts/convolution_benchmark.cu
index 70d397a..92accc4 100644
--- a/src/scripts/convolution_benchmark.cu
+++ b/src/scripts/convolution_benchmark.cu
@@ -217,7 +217,7 @@ int main(int argc, char* argv[]) {
 
     /*
     printf("Checking CUDA compatibility.\n");
-    bool cuda_compatible = check_cuda_compatibility();
+    bool cuda_compatible = cuda_setup(true);
     if (!cuda_compatible) {
         printf(RED "CUDA not compatible, skipping tests.\n" RESET);
         return 0;
diff --git a/test/cnn_convolution.cu b/test/cnn_convolution.cu
index d18c5c2..5dcba85 100644
--- a/test/cnn_convolution.cu
+++ b/test/cnn_convolution.cu
@@ -192,7 +192,7 @@ void run_convolution_test(int input_width, int output_width, int rows, int colum
 
 int main() {
     printf("Checking CUDA compatibility.\n");
-    bool cuda_compatible = check_cuda_compatibility();
+    bool cuda_compatible = cuda_setup(true);
     if (!cuda_compatible) {
         printf(RED "CUDA not compatible, skipping tests.\n" RESET);
         return 0;
diff --git a/test/cnn_function.cu b/test/cnn_function.cu
index 4d191f8..9ba01b2 100644
--- a/test/cnn_function.cu
+++ b/test/cnn_function.cu
@@ -91,7 +91,7 @@ void test1(int activation, bool use_local_kernel) {
 
 int main() {
     printf("Checking CUDA compatibility.\n");
-    bool cuda_compatible = check_cuda_compatibility();
+    bool cuda_compatible = cuda_setup(true);
     if (!cuda_compatible) {
         printf(RED "CUDA not compatible, skipping tests.\n" RESET);
         return 0;
diff --git a/test/cnn_matrix_multiplication.cu b/test/cnn_matrix_multiplication.cu
index e00a260..320cb2a 100644
--- a/test/cnn_matrix_multiplication.cu
+++ b/test/cnn_matrix_multiplication.cu
@@ -127,7 +127,7 @@ void run_matrices_test(int n, int p, int q) {
 
 int main() {
     printf("Checking CUDA compatibility.\n");
-    bool cuda_compatible = check_cuda_compatibility();
+    bool cuda_compatible = cuda_setup(true);
     if (!cuda_compatible) {
         printf(RED "CUDA not compatible, skipping tests.\n" RESET);
         return 0;
diff --git a/test/memory_management.cu b/test/memory_management.cu
index a575a55..d9161a7 100644
--- a/test/memory_management.cu
+++ b/test/memory_management.cu
@@ -18,7 +18,7 @@ __global__ void check_access(int* array, int range) {
 
 int main() {
     printf("Checking CUDA compatibility.\n");
-    bool cuda_compatible = check_cuda_compatibility();
+    bool cuda_compatible = cuda_setup(true);
     if (!cuda_compatible) {
         printf(RED "CUDA not compatible, skipping tests.\n" RESET);
         return 0;