1

I am making this object detector with TensorFlow 2 and I would like to get better performance because currently, I am getting about 0.09 fps! I heard that GPU is way faster than CPU but I have seemed to run into a problem. First I tried installing TensorFlow 2 GPU by doing pip install tensorflow-gpu==2.0.0-rc1 but when I ran my code it said:

2020-07-11 02:06:31.469406: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2020-07-11 02:06:31.475096: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cublas64_10.dll'; dlerror: cublas64_10.dll not found
2020-07-11 02:06:31.481287: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cufft64_10.dll'; dlerror: cufft64_10.dll not found
2020-07-11 02:06:31.486505: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'curand64_10.dll'; dlerror: curand64_10.dll not found
2020-07-11 02:06:31.492196: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cusolver64_10.dll'; dlerror: cusolver64_10.dll not found
2020-07-11 02:06:31.498517: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cusparse64_10.dll'; dlerror: cusparse64_10.dll not found
2020-07-11 02:06:31.503502: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudnn64_7.dll'; dlerror: cudnn64_7.dll not found

So I went to look and found I need to install something called Cuda but when I checked TensorFlow 2 only accepted NVIDIA® GPU card with CUDA® Compute Capability 3.5 or higher and if you look the GeForce GTX 770 has a 3.0. I have tried installing Cuda 10 but it keeps giving me errors. So what do I do? Do I have to downgrade Tensorflow or is there a workaround to getting TensorFlow 2 GPU for GeForce GTX 770? I am using Windows 10 btw.

Fanfer123
  • 67
  • 11
  • According to this (https://forums.developer.nvidia.com/t/gtx-770-and-compute-capability-3-5/29706/12), you might not be able to use cuda compute capability 3.5. In this case, you can downgrade your tensorflow version. Depending on your use case, you can use cloud services or if it's just for a short project/research, you can use free services like google colab – SajanGohil Jul 11 '20 at 16:05
  • @SajanGohil So how would I go about this? Do I first need to find a downgraded version then install it and then install Cuda 3.0? What is the latest version of TensorFlow that supports 3.0? – Fanfer123 Jul 11 '20 at 17:18
  • I am not sure, you will need to look at the the docs for it – SajanGohil Jul 11 '20 at 17:19
  • As per this Tensorflow Documentation (https://www.tensorflow.org/install/source_windows#gpu) and this Stack Overflow Answer (https://stackoverflow.com/a/59248949/11530462), you may have to downgrade `Tensorflow Version` to `1.12-GPU` as it supports `Compute Capability of 3.0`. Or you may have to upgrade your `GPU` to `GeForce GTX 780` or higher, as per this link (https://developer.nvidia.com/cuda-gpus#compute), to be able to work with `higher versions of Tensorflow`. Thanks! –  Jul 20 '20 at 11:56

0 Answers0