How to use the GPU server

Computer Center has three operational GPU servers with Nvidia Tesla A40, A100 and S1070 GPU cards.

• gpu01.cc.iitk.ac.in ( IP address: 172.31.1.83)
• gpu02.cc.iitk.ac.in ( IP address: 172.31.1.84)
• gpu03.cc.iitk.ac.in ( IP address: 172.31.1.85)
• gpu04.cc.iitk.ac.in ( IP address: 172.31.1.86)

As per present policy gpu01, gpu02, gpu03, and gpu04 servers can be accessed by any user with a shell account by logging in via ssh (X-11 forwarding enabled).

	# ssh <IITK-username>@<server-name> -X 
	e.g. ssh soumitri@gpu01.cc.iitk.ac.in -X

Users need to send the following information to rajany@iitk.ac.in to activate their user account.

1. Username
2. Supervisor Email ID.
3. A paragraph of work to be done on the GPUs.
4. Duration for which GPU is required.
5. Expected storage space to be used in the scratch directory.
6. If any additional library is required, that needs to be well informed in advance.
7. Undertaking that the GPU will only be used for academic work.

The following "hello world" program may help you in getting started with CUDA compiler to compile a program. Please note this program does not use cuda cores.

	$ vi helloworld.cu
	-------------------------------
	$include "stdio.h"
	int main()
		{
		printf("Hello, world\n");
		return 0;
		}
	-------------------------------
	$ nvcc -o helloworld helloworld.cu
	$./helloworld

With the above sample C program, CUDA compiler is used to compile it. Use nvcc to compile programs which has CUDA functionality.