Installation Verification

The following is the minimum startup example for running GPU tensor computation.

python3 - <<'PY'
import ctypes 
import paddle  

print("load_libcuda", ctypes.CDLL("libcuda.so.1")._name) 
print("paddle_version", paddle.__version__) 
print("compiled_with_cuda", paddle.device.is_compiled_with_cuda()) 
print("device_count", paddle.device.cuda.device_count())  

paddle.set_device("gpu:0") 
x = paddle.arange(6, dtype="float32").reshape([2, 3]) 
y = paddle.ones([3, 2], dtype="float32") 
z = paddle.matmul(x, y) + 1 
paddle.device.synchronize()  

print("place", z.place) 
print("result", z.numpy().tolist()) 
PY

The expected output is as follows.

load_libcuda libcuda.so.1 
paddle_version 3.3.0.dev20260319 
compiled_with_cuda True 
device_count 1 place 
Place(gpu:0) 
result [[4.0, 4.0], [13.0, 13.0]]

The pass criteria are as follows.

Paddle is successfully imported.
compiled_with_cuda is True.
The value of device_count is greater than 0.
The tensor runs on gpu:0.
The GPU tensor calculation result is correct.

Parent topic: Installation Guide