[SOLVED]Run get_start.py error


class ‘tvm.tensor.Tensor’>
[14:03:08] /home/xxx/tvm/src/codegen/opt/build_cuda_on.cc:75: cannot detect compute capability from your device, fall back to compute_30.
Traceback (most recent call last):
File “get_started.py”, line 127, in
a = tvm.nd.array(np.random.uniform(size=n).astype(A.dtype), ctx)
File “/home/xxx/tvm/python/tvm/ndarray.py”, line 199, in array
return empty(arr.shape, arr.dtype, ctx).copyfrom(arr)
File “/home/xxx/tvm/python/tvm/_ffi/ndarray.py”, line 113, in empty
File “/home/xxx/tvm/python/tvm/_ffi/base.py”, line 72, in check_call
raise TVMError(py_str(_LIB.TVMGetLastError()))
tvm._ffi.base.TVMError: [14:03:08] /home/xxx/tvm/src/runtime/cuda/cuda_device_api.cc:93: Check failed: e == cudaSuccess || e == cudaErrorCudartUnloading CUDA: CUDA driver version is insufficient for CUDA runtime version

Stack trace returned 10 entries:
[bt] (0) /home/xxx/tvm/build/libtvm.so(+0x18093d5) [0x7f631461d3d5]
[bt] (1) /home/xxx/tvm/build/libtvm.so(+0x1f78d71) [0x7f6314d8cd71]
[bt] (2) /home/xxx/tvm/build/libtvm.so(tvm::runtime::NDArray::Empty(std::vector<long, std::allocator >, DLDataType, DLContext)+0x1e6) [0x7f6314d43ff6]
[bt] (3) /home/xxx/tvm/build/libtvm.so(TVMArrayAlloc+0xe9) [0x7f6314d44149]
[bt] (4) /usr/lib/x86_64-linux-gnu/libffi.so.6(ffi_call_unix64+0x4c) [0x7f6363f93dae]
[bt] (5) /usr/lib/x86_64-linux-gnu/libffi.so.6(ffi_call+0x22f) [0x7f6363f9371f]
[bt] (6) /usr/lib/python2.7/lib-dynload/_ctypes.x86_64-linux-gnu.so(_ctypes_callproc+0x2a4) [0x7f63641a6b04]
[bt] (7) /usr/lib/python2.7/lib-dynload/_ctypes.x86_64-linux-gnu.so(+0x10505) [0x7f63641a6505]
[bt] (8) python(PyEval_EvalFrameEx+0x54a0) [0x560f53ce0420]
[bt] (9) python(PyEval_EvalCodeEx+0x6da) [0x560f53cd8d0a]

Has anyone encountered this problem?


What was your CUDA install process; are you able to run other CUDA programs? It seems the best solution would just be a fresh install of the NVIDIA drivers and CUDA to ensure the runtime and driver version match.


yes, I can run the cuda sample right.

I solve this problem after reinstall the driver and cuda. thank you.