RTX 2080: no kernel image is available for execution on the device

[Mickey Iluz]

Hi,

I built arrayfire v3.6.1 with CUDA 10 on Ubuntu 16.04. Running arrayfire code failed on the RTX 2080 (but succeeded when a GTX 1080 Ti was in that PC).

I managed to reproduce it with a short python example:

Python 2.7.12 (default, Dec  4 2017, 14:50:18) 

Type "copyright", "credits" or "license" for more information.

IPython 5.0.0 -- An enhanced Interactive Python.

?         -> Introduction and overview of IPython's features.

%quickref -> Quick reference.

help      -> Python's own help system.

object?   -> Details about 'object', use 'object??' for extra details.

In [1]: import arrayfire as af

In [2]: A = af.random.randu(5,5)

In [3]: A

Out[3]: 

arrayfire.Array()

Type: float

Dims: (5, 5)

In [4]: af.transpose(A)

---------------------------------------------------------------------------

RuntimeError                              Traceback (most recent call last)

<ipython-input-4-37131a6a6f59> in <module>()

----> 1 af.transpose(A)

/usr/local/lib/python2.7/dist-packages/arrayfire/array.pyc in transpose(a, conj)

    316     """

    317     out = Array()

--> 318     safe_call(backend.get().af_transpose(c_pointer(out.arr), a.arr, conj))

    319     return out

    320 

/usr/local/lib/python2.7/dist-packages/arrayfire/util.pyc in safe_call(af_error)

     77         err_len = c_dim_t(0)

     78         backend.get().af_get_last_error(c_pointer(err_str), c_pointer(err_len))

---> 79         raise RuntimeError(to_str(err_str))

     80 

     81 def get_version():

RuntimeError: In function void cuda::kernel::transpose(cuda::Param<T>, cuda::CParam<T>, int) [with T = float; bool conjugate = false]

In file src/backend/cuda/kernel/transpose.hpp:116

CUDA Error (48): no kernel image is available for execution on the device

Any ideas?

[Guillaume LAURENT]

Hi Mickey,

Which nvidia driver are you using and what is the result of af.info() ?

Cheers
Guillaume

--
You received this message because you are subscribed to the Google Groups "ArrayFire Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to arrayfire-use...@googlegroups.com.
To post to this group, send email to arrayfi...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/arrayfire-users/f2463d5e-2646-4cdb-9c0f-80b3499fb000%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Mickey Iluz]

The problem was that wrong CUDA compute was selected at the CMake stage (perhaps due to the fact that both 1080 Ti and RTX 2080 were connected at that time).

Running these solved the problem:

mkdir build 
cd build 
cmake -DCUDA_architecture_build_target:STRING=7.5 ..

(Thanks umar456!)