Ubuntu查看CUDA的CUDA_ARCH_BIN

我们在使用源码编译带cuda的opencv时，需要设置显卡的CUDA_ARCH_BIN，本文介绍一下获得该值的方法

方法一

安装好cuda之后，可以从cuda sample中获得

cd /usr/local/cuda/samples/1_Utilities/deviceQuery

sudo make -j8

./deviceQuery

如上，运行后输出信息如下

1. nvidia@nvidia-X10SRA:/usr/local/cuda/samples/1_Utilities/deviceQuery$ ./deviceQuery

2. ./deviceQuery Starting...

3. CUDA Device Query (Runtime API) version (CUDART static linking)

4. Detected 3 CUDA Capable device(s)

5. Device 0: "Tesla T4"

6. CUDA Driver Version / Runtime Version 10.2 / 10.2

7. CUDA Capability Major/Minor version number: 7.5

8. Total amount of global memory: 15110 MBytes (15843721216 bytes)

9. (40) Multiprocessors, ( 64) CUDA Cores/MP: 2560 CUDA Cores

10. GPU Max Clock rate: 1590 MHz (1.59 GHz)

11. Memory Clock rate: 5001 Mhz

12. Memory Bus Width: 256-bit

13. L2 Cache Size: 4194304 bytes

14. Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)

15. Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers

16. Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers

17. Total amount of constant memory: 65536 bytes

18. Total amount of shared memory per block: 49152 bytes

19. Total number of registers available per block: 65536

20. Warp size: 32

21. Maximum number of threads per multiprocessor: 1024

22. Maximum number of threads per block: 1024

23. Max dimension size of a thread block (x,y,z): (1024, 1024, 64)

24. Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)

25. Maximum memory pitch: 2147483647 bytes

26. Texture alignment: 512 bytes

27. Concurrent copy and kernel execution: Yes with 3 copy engine(s)

28. Run time limit on kernels: No

29. Integrated GPU sharing Host Memory: No

30. Support host page-locked memory mapping: Yes

31. Alignment requirement for Surfaces: Yes

32. Device has ECC support: Enabled

33. Device supports Unified Addressing (UVA): Yes

34. Device supports Compute Preemption: Yes

35. Supports Cooperative Kernel Launch: Yes

36. Supports MultiDevice Co-op Kernel Launch: Yes

37. Device PCI Domain ID / Bus ID / location ID: 0 / 2 / 0

38. Compute Mode:

39. < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

40. ……

可以看到的相关信息，包括cuda capability为7.5，2560个cuda核心等

方法二

如果安装cuda时没有安装samples，则可以使用下边的方法

 git clone https://github.com/NVIDIA-AI-IOT/deepstream_tlt_apps.git

cd deepstream_tlt_apps/TRT-OSS/x86

nvcc deviceQuery.cpp -o deviceQuery

./deviceQuery

同样会输出上边类似的信息

方法三

官网查询，选择自己对应的显卡型号查