Getting started

Users who wish to use Dirac must first submit a Request for Dirac Account form. This form includes fields to describe the proposed research. Users will be expected to provide a brief written summary of their results periodically.

Access to Dirac is via batch jobs only.  You first login to the Carver system:

ssh -l username carver.nersc.gov

and then submit batch jobs to Dirac.  See  Running Jobs on Dirac.

Running Your First CUDA Code: Vector Addition

Open a new file called vectorAdd.cu with a text editor such as emacs or vi. Paste the contents of the below code into the file.

//  Kernel definition, see also section 4.2.3 of Nvidia Cuda Programming Guide
       __global__  void vecAdd(float* A, float* B, float* C)
       {
          // threadIdx.x is a built-in variable  provided by CUDA at runtime
          int i = threadIdx.x;
          A[i]=0;
          B[i]=i;
          C[i] = A[i] + B[i];
       }

       #include  <stdio.h>
       #define  SIZE 10
       int  main()
       {
           int N=SIZE;
           float A[SIZE], B[SIZE], C[SIZE];
           float *devPtrA;
           float *devPtrB;
           float *devPtrC;
           int memsize= SIZE * sizeof(float);

           cudaMalloc((void**)&devPtrA, memsize);
           cudaMalloc((void**)&devPtrB, memsize);
           cudaMalloc((void**)&devPtrC, memsize);
           cudaMemcpy(devPtrA, A, memsize,  cudaMemcpyHostToDevice);
           cudaMemcpy(devPtrB, B, memsize,  cudaMemcpyHostToDevice);
           // __global__ functions are called:  Func<<< Dg, Db, Ns  >>>(parameter);
           vecAdd<<<1, N>>>(devPtrA,  devPtrB, devPtrC);
           cudaMemcpy(C, devPtrC, memsize,  cudaMemcpyDeviceToHost);

           for (int i=0; i<SIZE; i++)
            printf("C[%d]=%f\n",i,C[i]);

            cudaFree(devPtrA);
           cudaFree(devPtrA);
           cudaFree(devPtrA);
       }

Compile the Program

% module load cuda
% nvcc vectorAdd.cu

Submit Your Job to the Queue

The qsub command is used to submit job the Dirac compute nodes.

[virajp83@cvrsvc04 ~]$ qsub -I -V -q dirac_int -l nodes=1:ppn=8:fermi 
qsub: waiting for job 494318.cvrsvc09-ib to start
qsub: job 494318.cvrsvc09-ib ready

Run Your job

Your job will be placed on a Dirac compute node. Run it as follows

[virajp83@dirac41 ~]$ ./a.out 
C[0]=0.000000
C[1]=1.000000
C[2]=2.000000
C[3]=3.000000
C[4]=4.000000
C[5]=5.000000
C[6]=6.000000
C[7]=7.000000
C[8]=8.000000
C[9]=9.000000