经过探索，还是conda安装比较方便，可以安装miniconda，见另一篇文章。

1. 新建虚拟环境，要指定python3.7，之前3.8，3.9都走了弯路

conda create -n py37tf python=3.7

conda activae py37tf

2. 安装tensorflow

conda install tensorflow-gpu

3. 验证

Python 3.7.9 (default, Aug 31 2020, 12:42:55)
[GCC 7.3.0] :: Anaconda, Inc. on linux
>>> import tensorflow
>>> tensorflow.test.is_gpu_available()
WARNING:tensorflow:From <stdin>:1: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.config.list_physical_devices('GPU')` instead.
2020-12-08 18:00:20.073746: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2020-12-08 18:00:20.102014: I tensorflow/core/platform/profile_utils/cpu_utils.cc:102] CPU Frequency: 2599990000 Hz
2020-12-08 18:00:20.102510: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x558847e0a280 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-12-08 18:00:20.102533: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-12-08 18:00:20.103186: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-12-08 18:00:20.121878: 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
2020-12-08 18:00:20.122160: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1660 Ti computeCapability: 7.5
coreClock: 1.59GHz coreCount: 24 deviceMemorySize: 5.80GiB deviceMemoryBandwidth: 268.26GiB/s
2020-12-08 18:00:20.122284: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-12-08 18:00:20.123416: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-12-08 18:00:20.124460: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-12-08 18:00:20.124626: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-12-08 18:00:20.125812: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-12-08 18:00:20.126517: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-12-08 18:00:20.128898: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-12-08 18:00:20.128966: 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
2020-12-08 18:00:20.129256: 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
2020-12-08 18:00:20.129484: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2020-12-08 18:00:20.129509: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-12-08 18:00:20.183243: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-12-08 18:00:20.183268: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108]      0
2020-12-08 18:00:20.183273: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 0:   N
2020-12-08 18:00:20.183378: 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
2020-12-08 18:00:20.183641: 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
2020-12-08 18:00:20.183880: 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
2020-12-08 18:00:20.184108: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/device:GPU:0 with 5265 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1660 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5)
2020-12-08 18:00:20.185437: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x558849716c40 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-12-08 18:00:20.185449: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce GTX 1660 Ti, Compute Capability 7.5
True
>>>

OK