diff --git a/mindspore/ccsrc/runtime/device/gpu/distribution/collective_fake_init.cc b/mindspore/ccsrc/runtime/device/gpu/distribution/collective_fake_init.cc
index cd511277456..2258ccf4421 100644
--- a/mindspore/ccsrc/runtime/device/gpu/distribution/collective_fake_init.cc
+++ b/mindspore/ccsrc/runtime/device/gpu/distribution/collective_fake_init.cc
@@ -21,15 +21,13 @@ namespace mindspore {
 namespace device {
 namespace gpu {
 void CollectiveFakeInitializer::InitCollective() {
-  MS_LOG(EXCEPTION)
-    << "You are trying to call 'init('nccl')', but this MindSpore package is built without NCCL. Please download GPU "
-       "version of MindSpore and try again.";
+  MS_LOG(EXCEPTION) << "You are trying to call 'init('nccl')', Please check "
+                       "this MindSpore package is GPU version and built with NCCL.";
 }
 
 void CollectiveFakeInitializer::FinalizeCollective() {
-  MS_LOG(EXCEPTION)
-    << "You are trying to call 'init('nccl')', but this MindSpore package is built without NCCL. Please download GPU "
-       "version of MindSpore and try again.";
+  MS_LOG(EXCEPTION) << "You are trying to call 'init('nccl')', Please check "
+                       "this MindSpore package is GPU version and built with NCCL.";
 }
 }  // namespace gpu
 }  // namespace device
diff --git a/mindspore/ccsrc/runtime/device/gpu/distribution/collective_init.cc b/mindspore/ccsrc/runtime/device/gpu/distribution/collective_init.cc
index 137956a6a12..65d7c27f786 100644
--- a/mindspore/ccsrc/runtime/device/gpu/distribution/collective_init.cc
+++ b/mindspore/ccsrc/runtime/device/gpu/distribution/collective_init.cc
@@ -32,12 +32,14 @@ const void *CollectiveInitializer::collective_handle() const { return collective
 void CollectiveInitializer::InitCollective() {
   void *handle = dlopen("libgpu_collective.so", RTLD_LAZY);
   if (handle == nullptr) {
-    MS_LOG(EXCEPTION) << "Loading libgpu_collective.so failed. Many reasons could cause this:\n"
-                         "1.libgpu_collective.so is not found.\n"
-                         "2.NCCL is not found or the user-installed NCCL version installed is incompatible: MindSpore "
-                         "requires NCCL-2.7.6.\n"
-                         "3.OpenMPI is not found or the user-installed OpenMPI version is incompatible: MindSpore "
-                         "requires OpenMPI-4.0.3.\n";
+    MS_LOG(EXCEPTION)
+      << "Loading libgpu_collective.so failed. Many reasons could cause this:\n"
+         "1.libgpu_collective.so is not found, please check this MindSpore package is GPU version and built "
+         "with distributed feature.\n"
+         "2.NCCL is not found or the user-installed NCCL version installed is incompatible: MindSpore "
+         "requires NCCL-2.7.6.\n"
+         "3.OpenMPI is not found or the user-installed OpenMPI version is incompatible: MindSpore "
+         "requires OpenMPI-4.0.3.\n";
   }
   auto mpi_init_funcptr = reinterpret_cast<InitMPI>(dlsym(handle, "InitMPI"));
   MS_EXCEPTION_IF_NULL(mpi_init_funcptr);
diff --git a/mindspore/ccsrc/runtime/device/gpu/gpu_memory_allocator.cc b/mindspore/ccsrc/runtime/device/gpu/gpu_memory_allocator.cc
index 46a135ffffe..4aa48a1f402 100644
--- a/mindspore/ccsrc/runtime/device/gpu/gpu_memory_allocator.cc
+++ b/mindspore/ccsrc/runtime/device/gpu/gpu_memory_allocator.cc
@@ -37,8 +37,10 @@ bool GPUMemoryAllocator::Init() {
     MS_LOG(INFO) << "GPU device total memory size " << total_size << ", current free memory size " << free_size
                  << ", set max available memory size " << available_device_memory_ << ".";
   } else {
-    MS_LOG(EXCEPTION) << "GPU device memory error, total memory size " << total_size << ", current free memory size "
-                      << free_size << ", set max available memory size " << available_device_memory_ << ".";
+    MS_LOG(EXCEPTION)
+      << "The total size or free size or max_device_memory size of GPU memory can't be zero, total memory size "
+      << total_size << ", current free memory size " << free_size << ", set max available memory size "
+      << available_device_memory_ << ".";
   }
   return true;
 }
@@ -50,7 +52,7 @@ void GPUMemoryAllocator::CheckMaxDeviceMemory() const {
   //  Currently not support modifying the max device memory.
   if (limited_device_memory_ != max_device_memory) {
     MS_LOG(EXCEPTION)
-      << "Can't change context param max_device_memory in runtime, currently effective max_device_memory("
+      << "Can't change or set context param max_device_memory during running, currently effective max_device_memory("
       << limited_device_memory_ << "GB), set new max_device_memory(" << max_device_memory << "GB) failed.";
   }
 }
diff --git a/mindspore/ccsrc/runtime/device/gpu/kernel_info_setter.cc b/mindspore/ccsrc/runtime/device/gpu/kernel_info_setter.cc
index 7d62e65d10c..6d4ce0311e1 100644
--- a/mindspore/ccsrc/runtime/device/gpu/kernel_info_setter.cc
+++ b/mindspore/ccsrc/runtime/device/gpu/kernel_info_setter.cc
@@ -80,7 +80,7 @@ std::string SupportedTypeList(const CNodePtr &kernel_node) {
   std::string op_name = AnfAlgo::GetCNodeName(kernel_node);
   auto op_info_ptr = mindspore::kernel::OpLib::FindOp(op_name, kernel::OpImplyType::kAKG);
   if (op_info_ptr == nullptr) {
-    MS_LOG(EXCEPTION) << "Unsupported op [" << op_name << "]";
+    MS_LOG(EXCEPTION) << "Unsupported op [" << op_name << "] on GPU";
   }
   (void)ParseMetadata(kernel_node, op_info_ptr, kernel::Processor::CUDA, &kernel_info_list);
   for (size_t i = 0; i < kernel_info_list.size(); i++) {
@@ -135,7 +135,7 @@ bool SelectAkgKernel(const CNodePtr &kernel_node, const std::shared_ptr<KernelBu
                              return CheckKernelInfo(alternative_kernel_info, selected_kernel_info);
                            });
   if (!match) {
-    MS_LOG(ERROR) << "Not find op[" << op_name << "] in akg";
+    MS_LOG(ERROR) << "Not find op[" << op_name << "] which both match data type and format in akg";
     return false;
   }
   return true;