diff --git a/mindspore/ccsrc/runtime/device/gpu/gpu_kernel_runtime.cc b/mindspore/ccsrc/runtime/device/gpu/gpu_kernel_runtime.cc
index 207893834f..dbdd17c915 100644
--- a/mindspore/ccsrc/runtime/device/gpu/gpu_kernel_runtime.cc
+++ b/mindspore/ccsrc/runtime/device/gpu/gpu_kernel_runtime.cc
@@ -302,7 +302,7 @@ void GPUKernelRuntime::AllocInplaceNodeMemory(const session::KernelGraph *graph)
     auto output_size = kernel_mod->GetOutputSizeList();
     auto ret = mem_manager_->MallocMemFromMemPool(device_address, output_size[output_index]);
     if (!ret) {
-      MS_LOG(EXCEPTION) << "Cannot alloc address, tensor size is: " << output_size[output_index];
+      MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << output_size[output_index];
     }
 
     for (auto &node : item) {
@@ -662,8 +662,9 @@ bool GPUKernelRuntime::LaunchKernelDynamic(const session::KernelGraph *graph, bo
         if (profiler_inst->GetEnableFlag()) {
           profiler_inst->OpDataProducerBegin(kernel->fullname_with_scope(), stream_);
         }
-        CHECK_OP_RET_WITH_EXCEPT(kernel_mod->Launch(kernel_inputs, kernel_workspaces, kernel_outputs, stream_),
-                                 "Launch kernel failed.");
+        if (!kernel_mod->Launch(kernel_inputs, kernel_workspaces, kernel_outputs, stream_)) {
+          MS_LOG(EXCEPTION) << "Launch kernel failed: " << kernel->fullname_with_scope();
+        }
         if (profiler_inst->GetEnableFlag()) {
           profiler_inst->OpDataProducerEnd();
           if (profiler_inst->GetSyncEnableFlag()) {
diff --git a/mindspore/ccsrc/runtime/device/gpu/gpu_memory_manager.cc b/mindspore/ccsrc/runtime/device/gpu/gpu_memory_manager.cc
index 1a8da802d5..20e3323544 100644
--- a/mindspore/ccsrc/runtime/device/gpu/gpu_memory_manager.cc
+++ b/mindspore/ccsrc/runtime/device/gpu/gpu_memory_manager.cc
@@ -106,7 +106,9 @@ uint8_t *GPUMemoryManager::MallocStaticMem(size_t size, bool, uint32_t) {
   MS_EXCEPTION_IF_NULL(context_ptr);
   if (context_ptr->get_param<bool>(MS_CTX_ENABLE_DYNAMIC_MEM_POOL)) {
     auto device_ptr = MallocMemFromMemPool(size);
-    MS_EXCEPTION_IF_NULL(device_ptr);
+    if (device_ptr == nullptr) {
+      MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << size;
+    }
     return AddressOffset(device_ptr, 0);
   }
 
diff --git a/mindspore/ccsrc/runtime/device/kernel_runtime.cc b/mindspore/ccsrc/runtime/device/kernel_runtime.cc
index 238cdf5d81..26f3dfe26b 100644
--- a/mindspore/ccsrc/runtime/device/kernel_runtime.cc
+++ b/mindspore/ccsrc/runtime/device/kernel_runtime.cc
@@ -187,7 +187,7 @@ void KernelRuntime::RunOpAssignInputMemory(const std::vector<tensor::TensorPtr>
       MS_EXCEPTION_IF_NULL(mem_manager_);
       auto ret = mem_manager_->MallocMemFromMemPool(device_address, tensor_size);
       if (!ret) {
-        MS_LOG(EXCEPTION) << "Malloc device memory failed.";
+        MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << tensor_size;
       }
       AnfAlgo::SetOutputAddr(device_address, index, item.get());
     }
@@ -220,7 +220,7 @@ void KernelRuntime::RunOpAssignOutputMemory(const AnfNodePtr &kernel) {
     MS_EXCEPTION_IF_NULL(device_address);
     auto ret = mem_manager_->MallocMemFromMemPool(device_address, output_sizes[i]);
     if (!ret) {
-      MS_LOG(EXCEPTION) << "Malloc device memory failed.";
+      MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << output_sizes[i];
     }
     AnfAlgo::SetOutputAddr(device_address, i, kernel.get());
   }
@@ -238,7 +238,7 @@ void KernelRuntime::RunOpAssignWorkSpaceMemory(const AnfNodePtr &kernel) {
       MS_EXCEPTION_IF_NULL(device_address);
       auto ret = mem_manager_->MallocMemFromMemPool(device_address, workspace_lists[i]);
       if (!ret) {
-        MS_LOG(EXCEPTION) << "Malloc device memory failed.";
+        MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << workspace_lists[i];
       }
       AnfAlgo::SetWorkspaceAddr(device_address, i, kernel.get());
     }
@@ -651,7 +651,7 @@ void KernelRuntime::AssignValueNodeTensor(const ValueNodePtr &value_node, const
     MS_EXCEPTION_IF_NULL(address);
     if (ms_context->get_param<bool>(MS_CTX_ENABLE_PYNATIVE_INFER) &&
         !mem_manager_->MallocMemFromMemPool(address, node_size)) {
-      MS_LOG(EXCEPTION) << "Cannot alloc address from memory pool when tensor size is: " << node_size;
+      MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << node_size;
     } else if (mem_manager_->MallocMem(kStaticMem, node_size, address, graph_id) == nullptr) {
       MS_LOG(EXCEPTION) << "Cannot alloc address when flag is: " << kStaticMem << ", tensor size is: " << node_size;
     }
@@ -692,7 +692,7 @@ void KernelRuntime::AssignStaticMemoryValueNode(session::KernelGraph *graph) {
       MS_EXCEPTION_IF_NULL(address);
       if (ms_context->get_param<bool>(MS_CTX_ENABLE_PYNATIVE_INFER) &&
           !mem_manager_->MallocMemFromMemPool(address, tensor_size)) {
-        MS_LOG(EXCEPTION) << "Cannot alloc address from memory pool when tensor size is: " << tensor_size;
+        MS_LOG(EXCEPTION) << "Device memory isn't enough and alloc failed, alloc size:" << tensor_size;
       } else if (mem_manager_->MallocMem(kStaticMem, tensor_size, address, graph->graph_id()) == nullptr) {
         MS_LOG(EXCEPTION) << "Cannot alloc address when flag is: " << kStaticMem << ", tensor size is: " << tensor_size;
       }