EventSynchronize

    while(getline(fbase, line))
    {
        cudaEventSynchronize(finishedMemcpyEvent);
        if (!(fbase >> user >> movie >> day >> rating)){
            data_idx++;
            continue;
        }
        // process the line. Store indices instead of values
        buffer[data_idx % batchsize] = DataPoint(user - 1, movie - 1, rating);

        if((data_idx % batchsize == batchsize - 1)){
            cudaMemcpyAsync(d_data[i], inbuffers[i],
                batchsize * sizeof(DataPoint), cudaMemcpyHostToDevice, s[i]);
            cudaEventRecord(finishedMemcpyEvent, s[i]);
            // RUN KERNEL
            cudaTrain(d_data[i], pack,
                batchsize, nfactors, s[i]);
            i = (i + 1) % nstreams;
            if(VERBOSE)
                printf("Batch %6d finished\n", data_idx/batchsize);
            buffer = inbuffers[i];
        }
        data_idx++;
    }