Tensorflow 2.1 with CUDA10.2 warnings ..

Hi,

I upgraded my TensorFlow to 2.1 and my CUDA is 10.2 when I import Tensorflow in python3 I get the following warnings -

2020-01-10 15:27:40.101443: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libnvinfer.so.6’; dlerror: libnvinfer.so.6: cannot open shared object file: No such file or directory
2020-01-10 15:27:40.101504: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libnvinfer_plugin.so.6’; dlerror: libnvinfer_plugin.so.6: cannot open shared object file: No such file or directory
2020-01-10 15:27:40.101512: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:30] Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

Will have any influence on inference time for object detection.

Regards,
Suman

same here - any guidance?

The TRT library is required only if you are using TF-TRT to accelerate inference performance. If you are using only native TensorFlow inference, you do not need to install TRT. For more on what TF-TRT is and how to use it see https://docs.nvidia.com/deeplearning/frameworks/tf-trt-user-guide/index.html.

Hi, I have installed TensorFlow 2.1, CUDA 10.2 and TensorflowRT 7. I get the same issue.

±----------------------------------------------------------------------------+
| NVIDIA-SMI 440.44 Driver Version: 440.44 CUDA Version: 10.2 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 1650 Off | 00000000:01:00.0 Off | N/A |
| N/A 27C P8 3W / N/A | 464MiB / 3911MiB | 3% Default |
±------------------------------±---------------------±---------------------+

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1622 G /usr/lib/xorg/Xorg 189MiB |
| 0 1806 G /usr/bin/gnome-shell 98MiB |
| 0 2177 G …quest-channel-token=4814796753767024128 173MiB |
±----------------------------------------------------------------------------+

=====================
>>> print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices(‘GPU’)))
2020-01-15 14:15:20.295385: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-01-15 14:15:20.324956: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-15 14:15:20.325290: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1650 computeCapability: 7.5
coreClock: 1.56GHz coreCount: 16 deviceMemorySize: 3.82GiB deviceMemoryBandwidth: 119.24GiB/s
2020-01-15 14:15:20.325355: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libcudart.so.10.1’; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory
2020-01-15 14:15:20.326427: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-15 14:15:20.327470: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-15 14:15:20.327653: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-15 14:15:20.328955: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-15 14:15:20.329583: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-15 14:15:20.331974: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-15 14:15:20.331986: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices…
Num GPUs Available: 0

Assuming you are installing tensorflow from pip, you will need CUDA 10.1 and TRT 6 rather than CUDA 10.2 and TRT 7. See https://www.tensorflow.org/install/gpu

Okay. Thank you :)

Hi, can you tell me if tensorflow is well installed?
I want to use my gpu with keras, but I don’t know if I can with the missing package/library.

r@R:~$ source ./venv/bin/activate
(venv) r@R:~$ python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
2020-01-30 17:47:24.700460: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer.so.6'; dlerror: libnvinfer.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-30 17:47:24.700514: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer_plugin.so.6'; dlerror: libnvinfer_plugin.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-30 17:47:24.700522: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:30] Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.
2020-01-30 17:47:25.171459: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-01-30 17:47:25.189936: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-30 17:47:25.190413: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-30 17:47:25.190471: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-30 17:47:25.191381: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-30 17:47:25.192315: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-30 17:47:25.192450: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-30 17:47:25.193340: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-30 17:47:25.193835: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-30 17:47:25.195862: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-30 17:47:25.195873: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2020-01-30 17:47:25.196094: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-01-30 17:47:25.218282: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3593425000 Hz
2020-01-30 17:47:25.218808: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x56a45f0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-01-30 17:47:25.218828: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-01-30 17:47:25.283949: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-30 17:47:25.284336: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x56a67e0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-01-30 17:47:25.284357: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce RTX 2070 SUPER, Compute Capability 7.5
2020-01-30 17:47:25.284431: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-30 17:47:25.284439: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      
tf.Tensor(-182.68456, shape=(), dtype=float32)
(venv) r@R:~$ python3
Python 3.6.9 (default, Nov  7 2019, 10:44:02) 
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from tensorflow.python.client import device_lib
2020-01-30 17:49:52.770620: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer.so.6'; dlerror: libnvinfer.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-30 17:49:52.770676: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer_plugin.so.6'; dlerror: libnvinfer_plugin.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-30 17:49:52.770686: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:30] Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.
>>> print(device_lib.list_local_devices())
2020-01-30 17:49:56.383796: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-01-30 17:49:56.410355: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3593425000 Hz
2020-01-30 17:49:56.410886: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x4b42440 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-01-30 17:49:56.410906: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-01-30 17:49:56.413206: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-01-30 17:49:56.519774: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-30 17:49:56.520170: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x4baf690 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-01-30 17:49:56.520189: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce RTX 2070 SUPER, Compute Capability 7.5
2020-01-30 17:49:56.520337: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-30 17:49:56.520814: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-30 17:49:56.520906: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-30 17:49:56.522247: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-30 17:49:56.523558: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-30 17:49:56.523772: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-30 17:49:56.525146: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-30 17:49:56.525986: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-30 17:49:56.529086: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-30 17:49:56.529104: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2020-01-30 17:49:56.529124: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-30 17:49:56.529132: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-01-30 17:49:56.529139: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 5064229787380351532
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 12658538345896065301
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 10216805759268722875
physical_device_desc: "device: XLA_GPU device"
]

When I tried some of this https://stackoverflow.com/questions/38009682/how-to-tell-if-tensorflow-is-using-gpu-acceleration-from-inside-python-shell I got some errors…

Hope this article helps:

https://medium.com/@shravan007.c/tensorflow-2-1-running-on-gpu-ubuntu-18-04-76b0c8b0a25d

or try below:
sess = tf.compat.v1.Session(config=tf.ConfigProto(log_device_placement=True))

Hi shravan007.c,
I didn’t follow your tutotial, I followed this one https://medium.com/@aspiring1/installing-cuda-toolkit-10-0-and-cudnn-for-deep-learning-with-tensorflow-gpu-on-ubuntu-18-04-lts-f7e968b24c98, but it’s basically the same, except that you install TensorRT.

Following your tutorial, I tried your test to see whether my tensorflow is using my gpu or not and I got this:

>>> import tensorflow as tf#allow growth to take up minimal resources
>>> config = tf.compat.v1.ConfigProto()
>>> config.gpu_options.allow_growth = True
>>> sess = tf.compat.v1.Session(config=config)
2020-01-31 09:48:46.849694: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-01-31 09:48:46.879703: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3593470000 Hz
2020-01-31 09:48:46.880639: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55e8330 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-01-31 09:48:46.880658: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-01-31 09:48:46.883469: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-01-31 09:48:46.988694: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 09:48:46.989095: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x5653c90 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-01-31 09:48:46.989117: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce RTX 2070 SUPER, Compute Capability 7.5
2020-01-31 09:48:46.989293: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
[b]2020-01-31 09:48:46.989822: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5[/b]
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-31 09:48:46.989945: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-31 09:48:47.026280: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-31 09:48:47.045987: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-31 09:48:47.050478: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-31 09:48:47.088574: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-31 09:48:47.093583: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-31 09:48:47.156175: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
<b>2020-01-31 09:48:47.156196: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform</b>.
Skipping registering GPU devices...
2020-01-31 09:48:47.156227: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-31 09:48:47.156237: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-01-31 09:48:47.156246: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
>>>

I got some library missing, I don’t know if it matters?

As I’m using tensorflow 2.0 or 2.1, I used tf.compat.v1.ConfigProto instead of tf.ConfigProto, and this is the result (the returning code says kinda the same as the code above):

>>> sess = tf.compat.v1.Session(config= tf.compat.v1.ConfigProto(log_device_placement=True)) 
2020-01-31 09:56:05.010812: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 09:56:05.011381: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-31 09:56:05.011497: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64
2020-01-31 09:56:05.011519: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-31 09:56:05.011550: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-31 09:56:05.011566: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-31 09:56:05.011584: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-31 09:56:05.011600: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-31 09:56:05.011615: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-31 09:56:05.011622: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2020-01-31 09:56:05.011640: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-31 09:56:05.011649: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-01-31 09:56:05.011659: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
Device mapping:
/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device
/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device
2020-01-31 09:56:05.011704: I tensorflow/core/common_runtime/direct_session.cc:358] Device mapping:
/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device
/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device

>>>

Still the missing library but It’s maybe due to tensorflow 2.1 + the last version of every programs (cudnn, cuda , etc…), I don’t know, but if it does not matter, I would be very happy :D

libcudart.so.10.1 is missing because you appear to have CUDA 10.2 installed rather than 10.1, which is the required version when using Google’s pre-compiled binaries available from pip. You will want to install CUDA 10.1 and then point your LD_LIBRARY_PATH at /usr/local/cuda-10.1/lib64.

Hi nluehr,
thanks for answering,
if I don’t downgrade cuda, will I be able to use it without problem?
If I downgrade cuda, should I just downgrade by using a

sudo apt-get install cuda-10-1

,
Or should I uninstall everything from the bottom and reinstall everything (cuda, cuda toolkit, cudnn, tensorflow)?

Yes, for TF to use the GPU you will need to install the 10.1 toolkit. You should not need to change any other packages. Just make sure LD_LIBRARY_PATH points at the 10.1 install.

Hi, I reinstall cuda toolkit 10.1 but I don’t know if it’s good now because I still can’t load some library, see the code. The followgin code are the instructions from my first comment:

>>> import tensorflow as tf#allow growth to take up minimal resources
>>> config = tf.compat.v1.ConfigProto()
>>> config.gpu_options.allow_growth = True
>>> sess = tf.compat.v1.Session(config=config)
2020-01-31 22:33:23.226699: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:23.227293: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-31 22:33:23.227334: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-01-31 22:33:23.227352: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-31 22:33:23.227369: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-31 22:33:23.227385: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-31 22:33:23.227401: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-31 22:33:23.227417: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-31 22:33:23.227433: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-31 22:33:23.227515: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:23.228105: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:23.228628: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-01-31 22:33:23.228656: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-31 22:33:23.228666: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-01-31 22:33:23.228673: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
2020-01-31 22:33:23.228779: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:23.229389: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:23.229927: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7188 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2070 SUPER, pci bus id: 0000:08:00.0, compute capability: 7.5)
>>> sess = tf.compat.v1.Session(config= tf.compat.v1.ConfigProto(log_device_placement=True)) 
2020-01-31 22:33:31.602586: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:31.603189: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-31 22:33:31.603231: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-01-31 22:33:31.603251: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-31 22:33:31.603268: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-31 22:33:31.603286: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-31 22:33:31.603301: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-31 22:33:31.603314: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-31 22:33:31.603327: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-31 22:33:31.603404: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:31.603992: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:31.604512: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-01-31 22:33:31.604537: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-31 22:33:31.604546: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-01-31 22:33:31.604552: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
2020-01-31 22:33:31.604652: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:31.605252: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:33:31.605784: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7188 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2070 SUPER, pci bus id: 0000:08:00.0, compute capability: 7.5)
Device mapping:
/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device
/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce RTX 2070 SUPER, pci bus id: 0000:08:00.0, compute capability: 7.5
2020-01-31 22:33:31.605839: I tensorflow/core/common_runtime/direct_session.cc:358] Device mapping:
/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device
/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce RTX 2070 SUPER, pci bus id: 0000:08:00.0, compute capability: 7.5

Then I got this:

r@R:~$ python3
Python 3.6.9 (default, Nov  7 2019, 10:44:02) 
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from tensorflow.python.client import device_lib
[b]2020-01-31 22:31:37.915221: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer.so.6'; dlerror: libnvinfer.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.1/lib64
2020-01-31 22:31:37.915278: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer_plugin.so.6'; dlerror: libnvinfer_plugin.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.1/lib64
2020-01-31 22:31:37.915286: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:30] Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.[/b]
>>> print(device_lib.list_local_devices())
2020-01-31 22:31:43.338413: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-01-31 22:31:43.369359: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3593170000 Hz
2020-01-31 22:31:43.369928: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x4aabf80 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-01-31 22:31:43.369950: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-01-31 22:31:43.372125: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-01-31 22:31:43.478569: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:31:43.478954: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x4b194d0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-01-31 22:31:43.478965: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce RTX 2070 SUPER, Compute Capability 7.5
2020-01-31 22:31:43.479064: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:31:43.479371: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:08:00.0 name: GeForce RTX 2070 SUPER computeCapability: 7.5
coreClock: 1.8GHz coreCount: 40 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 417.29GiB/s
2020-01-31 22:31:43.479507: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-01-31 22:31:43.480295: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-01-31 22:31:43.480948: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-01-31 22:31:43.481082: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-01-31 22:31:43.481922: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-01-31 22:31:43.482554: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-01-31 22:31:43.484603: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-01-31 22:31:43.484681: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:31:43.485034: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:31:43.485335: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-01-31 22:31:43.485356: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-01-31 22:31:43.486002: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-31 22:31:43.486016: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-01-31 22:31:43.486023: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
2020-01-31 22:31:43.486148: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:31:43.486724: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-01-31 22:31:43.487258: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/device:GPU:0 with 7188 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2070 SUPER, pci bus id: 0000:08:00.0, compute capability: 7.5)
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 12417720970126384552
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 281614650356661556
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 11379430570946171007
physical_device_desc: "device: XLA_GPU device"
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 7537721344
locality {
  bus_id: 1
  links {
  }
}
incarnation: 2375214635386360717
physical_device_desc: "device: 0, name: GeForce RTX 2070 SUPER, pci bus id: 0000:08:00.0, compute capability: 7.5"
]
r@R:~$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Fri_Feb__8_19:08:17_PST_2019
Cuda compilation tools, release 10.1, V10.1.105

But when I do nvidia-smi:

r@R:~$ nvidia-smi 
Fri Jan 31 22:38:12 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.33.01    Driver Version: 440.33.01    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 207...  On   | 00000000:08:00.0  On |                  N/A |
|  0%   32C    P8    17W / 215W |    322MiB /  7979MiB |      3%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1201      G   /usr/lib/xorg/Xorg                            18MiB |
|    0      1289      G   /usr/bin/gnome-shell                          49MiB |
|    0      1638      G   /usr/lib/xorg/Xorg                           128MiB |
|    0      1783      G   /usr/bin/gnome-shell                         122MiB |
+-----------------------------------------------------------------------------+
r@R:~$

CUDA Version: 10.2 vs 10.1 with nvcc --version.

Is it a problem?
I’m still missing some package.

I just reinstall cuda toolkit and tensorflow (because I moved from the virtual environment), I didn’t touch cudnn.

Hi,

I had got the same issue. When I was installing.
2020-01-31 09:56:05.011497: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libcudart.so.10.1’; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64

I was able to resolve it by reinstalling all the software. I actually did reinstall Ubuntu for this.
Just make sure when reinstallation is done. CUDA 10.1 is installed.

The libnvinfer libraries are only needed if you are intending to use NVIDIA’s TensorRT. This would require some changes to your model scripts. See https://docs.nvidia.com/deeplearning/frameworks/tf-trt-user-guide/index.html for details.

In your case, I expect you can safely ignore the remaining warning. If you want to eliminate them, install TensorRT version 6, which you can obtain from https://developer.nvidia.com/nvidia-tensorrt-download.

2020-07-03 08:35:16.689017: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1

Is this some kind of error? because i cannot run the model training after this msg