Skip to content

Device_Configuration.md

Latest commit

 

History

History
91 lines (69 loc) · 2.52 KB

Device_Configuration.md

File metadata and controls

91 lines (69 loc) · 2.52 KB

Device configuration

A GPU's configuration can be determined by reading information from cudaDeviceProp. This object is accessible from a CUDA runtime running on a system with Nvidia GPU.

The source code is here.

Instruction

To run the source code:

nvcc -o cuda-device-profile cuda-device-profile.cu

This should generate an excuble:

cuda-device-profile

in the same directory.

Now run the executable:

./cuda-device-profile

and the output should be similar to this:

Device count: 1
Name: NVIDIA GeForce RTX 4060 Laptop GPU
Compute capability: 8.9
##### MEMORY INFO FOR DEVICE 0 ---
Total global memory: 8328511488
Total constant memory: 65536
##### MP INFORMATION FOR DEVICE 0 ---
Multiprocessor count: 24
Max threads per block: 1024
Max threads dimensions: (1024, 1024, 64)
Max grid dimensions: (2147483647, 65535, 65535)https://developer.nvidia.com/cuda-gpus

The os system is:

casablanca@casablanca-xps:~/Documents/projects/cuda/github/CUDA-examples$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 22.04.3 LTS
Release:        22.04
Codename:       jammy

CUDA Compiler

The command to determine the compiler used for the example is nvcc --version.

Following is my output:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Aug_15_22:02:13_PDT_2023
Cuda compilation tools, release 12.2, V12.2.140
Build cuda_12.2.r12.2/compiler.33191640_0

Notes

A cudaDeviceProp object is defined by:

cudaDeviceProp prop;

A comprehensive list of attributes accessible by this object is already mentioned.

Some of important attributes are shown in this example:

  1. Device info: count, name.
  2. Max number of threads per block: maxThreadsPerBlock.
  3. Max thread dimensions: maxThreadsDim.
  4. Max grid dimensions: maxGridSize.

Blocks and threads info

Information about blocks and threads are important for setting up how a function will be parallelized in GPU.

In function invocation or kernel definition, we will see this notation:

<<<P, Q>>>

where
P denotes number of blocks to run in parallel;
Q denotes number of threads in each block.

The system in this example is from a laptop equipped with Nvidia RTX4060 GPU. Compute capability index for each GPU series is published here. You may compare it to your GPU output.