Hello all,
I am testing some sample CUDA functions to analyse the GPU performance of jetson Xavier NX device. Here is the code that I am executing in jupyterlab notebook.
import numpy as np
from numba import cuda
@cuda.jit
def add_gpu(x, out):
idx = cuda.grid(1)
out[idx] = x[idx] + 2
a = np.arange(10,dtype=np.float32)
d_a = cuda.to_device(a)
d_out = cuda.device_array_like(d_a)
nbr_block_per_grid = 2
nbr_thread_per_block = 5
add_gpu[nbr_block_per_grid, nbr_thread_per_block](d_a, d_out)
out = d_out.copy_to_host()
def add_cpu(x):
for i in range(x.size):
x[i]+=2
return x
%timeit add_gpu[nbr_block_per_grid, nbr_thread_per_block](d_a, d_out)
%timeit add_cpu(a)
This code worked fine when I tried in CUDA 10.2 but I upgraded the CUDA version from 10.2 to 11.2 recently. Then I tried to run the same code in 11.2 and now I am getting an error as below.
NvvmSupportError: No supported GPU compute capabilities found. Please check your cudatoolkit version matches your CUDA version.
This is the output that I got for “nvcc -V” command.
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Mon_Nov_30_19:09:42_PST_2020
Cuda compilation tools, release 11.2, V11.2.67
Build cuda_11.2.r11.2/compiler.29373293_0
Is this error occurring because of the CUDA version upgrade? Any solution to overcome from this?
Appreciate if anyone can help on this.