What does the computing power wrt CUDA mean?

Question

I am new to CUDA programming and know little about it. Could you tell me what “the ability to compute CUDA” means? When I use the following code on my university server, it showed me the following result.

for (device = 0; device < deviceCount; ++device) { cudaDeviceProp deviceProp; cudaGetDeviceProperties(&deviceProp, device); printf("\nDevice %d has compute capability %d.%d.\n", device, deviceProp.major, deviceProp.minor); } 

RESULT:

 Device 0 has compute capability 4199672.0. Device 1 has compute capability 4199672.0. Device 2 has compute capability 4199672.0. . . 

cudaGetDeviceProperties returns two main and minor fields. Could you tell me what this means 4199672.0. ?

Answer 1

Computing ability is a “set of functions” (both hardware and software functions) of a device. You may have heard the names of the NVIDIA GPU architecture “Tesla,” “Fermi,” or “Kepler.” Each of these architectures has features that may not be available in previous versions.

In the CUDA toolkit installation folder on your hard drive, find the file CUDA_C_Programming_Guide.pdf (or google it) and find Appendix F.1 . It describes the differences in capabilities of various computing capabilities.

Answer 2

As @dialer mentioned, computational ability is a collection of CUDA-related computing devices. As the NVidia CUDA API evolves, the number of Compute Capability increases. At the time of writing, NVidia's latest GPUs were Compute Capability 3.5. You can get some information about what differences mean by looking at this table on Wikipedia.

As @aland points out, your call probably failed, and what you get is the result using an uninitialized variable . You must wrap your call to cudaGetDeviceProps() with an error checking function or a macro call; cm.

What is the canonical way to check for errors using the CUDA API?

to discuss the best way to do this.

Answer 3

If you know which GPU you have, you can see a list of the computing capabilities of nvidia products here