Difference between revisions of "Applications/Cuda"

From HPC
Jump to: navigation , search
m
m (Batch Job)
 
(12 intermediate revisions by the same user not shown)
Line 1: Line 1:
 +
__TOC__
 +
 
==Application Details==
 
==Application Details==
  
* Description: CUDA is NVIDIA’s parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of the GPU (graphics processing unit).
+
* Description: CUDA is NVIDIA’s parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of the GPU (Graphics Processing Unit).
* Version: 6.5.14, 7.5.18, 8.0.61 and 9.0.176
+
* Version: 9.0.176,10.1.168 and 11.5.0 ('''preferred''')
* Modules: cuda/6.5.14, cuda/7.5.18,  cuda/8.0.61 and cuda/9.0.176
+
* Modules: cuda/9.0.176, and cuda/10.1.168 and cuda/11.5.0
* Licence: Free, but owned by NVidia  
+
* Licence: Free to download, but owned by NVidia  
 +
 
 +
'''Note''' : Version 8.0.61 is designated for retirement
  
 
==Usage Examples==
 
==Usage Examples==
 
{|
 
{|
 
|style="width:5%; border-width: 0" | [[File:icon_tick.png]]
 
|style="width:5%; border-width: 0" | [[File:icon_tick.png]]
|style="width:95%; border-width: 0" | NVIDIA CUDA® modules 7.5.18, 8.0.61 and 9.0.176 have the Deep Neural Network library (cuDNN) included. It is a GPU-accelerated library of primitives for deep neural networks. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers. cuDNN is part of the NVIDIA Deep Learning SDK.
+
|style="width:95%; border-width: 0" | All NVIDIA CUDA® modules have the Deep Neural Network library (cuDNN) included. It is a GPU-accelerated library of primitives for deep neural networks. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers. cuDNN is part of the NVIDIA Deep Learning SDK.
 
|-
 
|-
 
|}
 
|}
'''Note''': this example is done on a node with a GPU accelerator, usually access would be achieved with the scheduler
+
 
 +
 
 +
===Interactive Session===
 +
Running a CUDA-based program on a GPU node in an interactive session:
  
 
<pre style="background-color: black; color: white; border: 2px solid black; font-family: monospace, sans-serif;">
 
<pre style="background-color: black; color: white; border: 2px solid black; font-family: monospace, sans-serif;">
Line 21: Line 28:
 
Last login: Fri Mar 16 10:05:54 2018 from gpu03
 
Last login: Fri Mar 16 10:05:54 2018 from gpu03
  
[username@gpu03 ~]$ module add cuda/9.0.176
+
[username@gpu03 ~]$ module add cuda/11.5.0
 
[username@gpu03 ~]$ ./gpuTEST
 
[username@gpu03 ~]$ ./gpuTEST
  
 
</pre>
 
</pre>
  
==Further Information==
+
===Batch Job===
 +
 
 +
Running a CUDA-based program on a GPU node as a batch script below:
 +
 
 +
<pre style="font-family: monospace, sans-serif;">
 +
#!/bin/bash
 +
#SBATCH -J gpu-cuda
 +
#SBATCH -N 1
 +
#SBATCH --ntasks-per-node 1
 +
#SBATCH -D /home/user/
 +
#SBATCH -o %N.%j.%a.out
 +
#SBATCH -e %N.%j.%a.err
 +
#SBATCH -p gpu
 +
#SBATCH --exclusive
 +
 
 +
module load cuda/11.0.5
 +
 
 +
/home/user/gpu_program
 +
 
 +
</pre>
 +
 
 +
==Next Steps==
  
 
* [[Programming/Cuda|CUDA Programming Support]]
 
* [[Programming/Cuda|CUDA Programming Support]]
 
* [https://developer.nvidia.com/accelerated-computing https://developer.nvidia.com/accelerated-computing]
 
* [https://developer.nvidia.com/accelerated-computing https://developer.nvidia.com/accelerated-computing]
  
 
+
{{Modulepagenav}}
{|
 
|style="width:5%; border-width: 0" | [[File:icon_home.png]]
 
|style="width:95%; border-width: 0" |
 
* [[Main_Page|Home]]
 
* [[Applications|Application support]]
 
* [[General|General]]
 
* [[Training|Training]]
 
* [[Programming|Programming support]]
 
|-
 
|}
 

Latest revision as of 14:25, 17 November 2022

Application Details

  • Description: CUDA is NVIDIA’s parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of the GPU (Graphics Processing Unit).
  • Version: 9.0.176,10.1.168 and 11.5.0 (preferred)
  • Modules: cuda/9.0.176, and cuda/10.1.168 and cuda/11.5.0
  • Licence: Free to download, but owned by NVidia

Note : Version 8.0.61 is designated for retirement

Usage Examples

Icon tick.png All NVIDIA CUDA® modules have the Deep Neural Network library (cuDNN) included. It is a GPU-accelerated library of primitives for deep neural networks. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers. cuDNN is part of the NVIDIA Deep Learning SDK.


Interactive Session

Running a CUDA-based program on a GPU node in an interactive session:


[username@login01 ~]$ interactive -pgpu
salloc: Granted job allocation 1014031
Job ID 1014031 connecting to gpu03, please wait...
Last login: Fri Mar 16 10:05:54 2018 from gpu03

[username@gpu03 ~]$ module add cuda/11.5.0
[username@gpu03 ~]$ ./gpuTEST

Batch Job

Running a CUDA-based program on a GPU node as a batch script below:

#!/bin/bash
#SBATCH -J gpu-cuda
#SBATCH -N 1
#SBATCH --ntasks-per-node 1
#SBATCH -D /home/user/
#SBATCH -o %N.%j.%a.out
#SBATCH -e %N.%j.%a.err
#SBATCH -p gpu
#SBATCH --exclusive

module load cuda/11.0.5

/home/user/gpu_program

Next Steps





Modules | Main Page | Further Topics