MPI doesn't use Infiniband with CycleCloud

Question

I created a two node Cluster with the help of the cycle Cloud GUI provided by Microsoft. I used the slurm scheduler (version: 20.11.4-1), HC44rs instances for the two nodes and a D12_v12 for the master / scheduler node. The same OpenLogicCentOS-HPC:8_1:latest OS was used on all the nodes (CentOS 8).
I then tried to run a simple MPI ping-pong program to check the connection speed. As I only got around 900 Mpbs I assume that the nodes didn't communicate over the provided Infiniband connection. (For a packet size of 8 MiB i got a latency of around 0.156 s. To run the program I used "sbatch -N2 --wrap="mpirun -n 2 my_program"". I also tried some -mca options but nothing helped to solve my problem.

Based on this two articles by Microsoft I assumed that the Infiniband drivers are installed and are ready to be used with any MPI library (https://learn.microsoft.com/en-us/azure/virtual-machines/workloads/hpc/enable-infiniband, https://learn.microsoft.com/en-us/azure/virtual-machines/workloads/hpc/setup-mpi). As suggested in the article I first tried HPC-X and then OpenMPI.
I also ran ifconfig to ensure that the Infiniband interface actaully exists and with ethtool I checked the speed it advertises. The following screenshot shows the results taken on one of the nodes.

This is the code I used:

int ping_pong(long int message_size, int repetitions, FILE *output){  
        int my_rank, num_procs;  
        MPI_Init(NULL, NULL);  
        MPI_Comm_size(MPI_COMM_WORLD, &num_procs);  
        MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);  
        MPI_Status stat;  
  
        // allocate buffer  
        uint8_t *buf = (uint8_t*)malloc(message_size*sizeof(*buf));  
        // set values in buffer to zero  
        for(int i=0; i<message_size*sizeof(*buf); i++){  
    	buf[i] = (uint8_t)0;  
        }  
  
        int tag1 = 42;  
        int tag2 = 43;  
        double target_time = 5*60; // 5 minutes (in secs)  
        double my_time = 0;  
  
        //write header of csv file  
        if(my_rank==0){  
            fprintf(output, "stress test - #repetitions: %d- message size [B]: %ld\n", repetitions, message_size);  
        }  
  
        double start, end, elapsed_time;  
        int finished = 1;  
        int iter = 0;  
        while(finished){  
        iter++;  
        if(iter%1000==0){  
                printf("rank: %d, time: %f\n", my_rank, my_time);  
                fflush(stdout);  
        }  
  
        start = MPI_Wtime();              
        if(my_rank == 0){  
            MPI_Send(buf, message_size*sizeof(*buf), MPI_UINT8_T, 1, tag1, MPI_COMM_WORLD);  
            MPI_Recv(buf, message_size*sizeof(*buf), MPI_UINT8_T, 1, tag2, MPI_COMM_WORLD, &stat);  
        } else{ // rank == 1  
            MPI_Recv(buf, message_size*sizeof(*buf), MPI_UINT8_T, 0, MPI_ANY_TAG, MPI_COMM_WORLD, &stat);  
            if(stat.MPI_TAG==0){  
                finished=0;  
            }else{  
                MPI_Send(buf, message_size*sizeof(*buf), MPI_UINT8_T, 0, tag2, MPI_COMM_WORLD);  
            }  
       }  
  
        end = MPI_Wtime();  
        elapsed_time = end - start;  
        my_time += elapsed_time;  
        if(my_rank==0){  
            fprintf(output, "%f\n", elapsed_time);  
            if(my_time>target_time){  
                finished = 0;  
                int tag = 0;  
                MPI_Send(buf, message_size*sizeof(*buf), MPI_UINT8_T, 1, tag, MPI_COMM_WORLD);  
            }  
        }  
    }         
  
    free(buf);  
    MPI_Finalize();  
    return 0;  
    }  

Any help is appreciated, thanks in advance. If you need any further information or clarification I'd be happy to give it to you.