Cuda10.0 RTX3090 Value 'sm_86' is not defined for option 'gpu-name' tensorflow-gpu==1.15.2

Accelerated Computing CUDA CUDA Programming and Performance
1

Need help
tensorflow.python.framework.errors_impl.InternalError: 2 root error(s) found.
(0) Internal: ptxas exited with non-zero error code 65280, output: ptxas fatal : Value ‘sm_86’ is not defined for option ‘gpu-name’

 [[{{node cluster_1_1/xla_compile}}]]

(1) Internal: ptxas exited with non-zero error code 65280, output: ptxas fatal : Value ‘sm_86’ is not defined for option ‘gpu-name’

 [[{{node cluster_1_1/xla_compile}}]]
 [[cluster_1_1/merge_oidx_0/_23]]

0 successful operations.
0 derived errors ignored.

2

That’s correct, CUDA 10.x doesn’t support targeting any Ampere architecture.

You could build your own TF against CUDA 11.1
You could use a TF nightly with CUDA 11.1
You could use NGC
There may be other options.
I won’t be able to give you recipes.

3

thank you