gcc82的C++服务：
预测库：https://irepo.baidu-int.com/rest/prod/v3/baidu/vis-open/paddle/releases/1.8.4.1/files
paddle1.8.4
cuda10.1
libpaddle_fluid_gpu_cudnn7.so
libcudnn.so.7.5.0

process_num=1时多并发请求不会报错，但是process_num>1时经常会出core

改为cudn10.2 cudnn7.6也一样报错
export LD_LIBRARY_PATH=/usr/lib64/:./so/:/home/work/cuda-10.2/lib64/:/home/work/cudnn/cudnn_v7.6/cuda/lib64/

报错日志：
I0915 18:53:13.308756 10415 predictor.cpp:113] Predictor::predict
I0915 18:53:13.308779 10415 predictor.cpp:28] Predictor::feed
W0915 18:53:13.308849 10415 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 61, Driver API Version: 10.2, Runtime API Version: 10.1
W0915 18:53:13.308980 10415 device_context.cc:260] device: 0, cuDNN Version: 7.6.
find fluid handle
width224 height 224 channels 3
I0915 18:53:13.316826 10414 predictor.cpp:113] Predictor::predict
I0915 18:53:13.316859 10414 predictor.cpp:28] Predictor::feed
terminate called after throwing an instance of 'paddle::platform::EnforceNotMet'
what():

C++ Call Stacks (More useful to developers):

0 std::string paddle::platform::GetTraceBackString<char const*>(char const*&&, char const*, int)
1 paddle::platform::EnforceNotMet::EnforceNotMet(std::__exception_ptr::exception_ptr, char const*, int)
2 paddle::platform::TensorDescriptor::set(paddle::framework::Tensor const&, cudnnTensorFormat_t)
3 paddle::operators::CUDNNConvOpKernel::Compute(paddle::framework::ExecutionContext const&) const
4 std::_Function_handler<void (paddle::framework::ExecutionContext const&), paddle::framework::OpKernelRegistrarFunctor<paddle::platform::CUDAPlace, false, 0ul, paddle::operators::CUDNNConvOpKernel, paddle::operators::CUDNNConvOpKernel, paddle::operators::CUDNNConvOpKernelpaddle::platform::float16 >::operator()(char const*, char const*, int) const::{lambda(paddle::framework::ExecutionContext const&)#1}>::_M_invoke(std::_Any_data const&, paddle::framework::ExecutionContext const&)
5 paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&, paddle::framework::RuntimeContext*) const
6 paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&) const
7 paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, paddle::platform::Place const&)
8 paddle::framework::NaiveExecutor::Run()
9 paddle::AnalysisPredictor::ZeroCopyRun()

Python Call Stacks (More useful to users):

File "/home/ssd9/nizihan/env/python-gcc482-paddle-1.7.1/lib/python2.7/site-packages/paddle/fluid/framework.py", line 2525, in append_op
attrs=kwargs.get("attrs", None))
File "/home/ssd9/nizihan/env/python-gcc482-paddle-1.7.1/lib/python2.7/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op
return self.main_program.current_block().append_op(*args,**kwargs)
File "/home/ssd9/nizihan/env/python-gcc482-paddle-1.7.1/lib/python2.7/site-packages/paddle/fluid/layers/nn.py", line 1405, in conv2d
"data_format": data_format,
File "/home/ssd9/nizihan/plate/train_code/plate_type_classification_degree/models/resnet_vd.py", line 146, in conv_bn_layer
bias_attr=False)
File "/home/ssd9/nizihan/plate/train_code/plate_type_classification_degree/models/resnet_vd.py", line 67, in net
name='conv1_1')
File "infer.py", line 100, in infer
out = model.net(input=image, class_dim=args.class_dim)
File "infer.py", line 204, in main
infer(args)
File "infer.py", line 208, in
main()

Error Message Summary:

ExternalError: Cudnn error, CUDNN_STATUS_BAD_PARAM at (/home/scmbuild/workspaces_cluster.dev/baidu.lib.paddlepaddle/baidu/lib/paddlepaddle/Paddle/paddle/fluid/platform/cudnn_desc.h:155)
[operator < conv2d > error]