S = np.array2string(self._value, prefix=prefix, suffix=',',įile '/home/holl/miniconda3/envs/phiflow2_tf/lib/python3.8/site-packages/jax/interpreters/xla.py', line 1122, in _value 21:26:30.353849: E external/org_tensorflow/tensorflow/compiler/xla/pjrt/pjrt_stream_executor_:1881] Execution of replica 0 failed: Internal: Could not find the corresponding functionįile '/home/holl/miniconda3/envs/phiflow2_tf/lib/python3.8/site-packages/jax/interpreters/xla.py', line 1181, in _repr_ 21:26:30.353808: E external/org_tensorflow/tensorflow/stream_executor/cuda/cuda_:682] failed to get PTX kernel 'broadcast_2' from module: CUDA_ERROR_NOT_FOUND: named symbol not found 21:26:30.353307: W external/org_tensorflow/tensorflow/stream_executor/gpu/asm_:194] Used ptxas at /usr/local/cuda-11.0/bin/ptxas 21:26:30.353284: W external/org_tensorflow/tensorflow/stream_executor/gpu/asm_:191] Falling back to the CUDA driver for PTX compilation ptxas does not support CC 8.6 I can import Jax from Python but the first operation throws an error. I installed Jax into my Python 3.8.6 conda environment by running pip3 install -upgrade jax jaxlib=0.1.62+cuda110 -f The path /usr/local/cuda points to that installation.

I have a GeForce RTX 3090 with CUDA 11.0 installed on Ubuntu 16.04 and the installation works fine with TensorFlow.

Install cuda on ubuntu 16.04