CUDA nvlink warning : SM Arch ('sm_35') not found - c++

Question

yesterday installed cuda-6.5 to my ubunutu14.04. I followed the steps stated in cuda's getting started guide. Checked for System requirements and mine was OK. Did pre-installations, uninstalled previously installed cuda, and installed package manager installation. All these steps were successfully performed. I skipped steps runfile installation and croos-build environment fro arm. In [post-installation actions][2] step, added
export PATH=/usr/local/cuda-6.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-6.5/lib64:$LD_LIBRARY_PATH
these lines to .profile file. I upgraded my driver to the latest available driver by running the command sudo apt-get install cuda-drivers. Also verified that I installed correct driver. Rebooted my computer and vaulla cuda-6.5 is build successfully. But when I compile my simpleCuda.cu file
#include <stdio.h>
#include <cuda.h>
#include<iostream>
#include <thrust/device_vector.h>
#include <thrust/logical.h>
#include <thrust/functional.h>
#include <cassert>
#include <cublas_v2.h>
using namespace std;
int main(){
float* dev,host;
cudaError_t stat = cudaMalloc((void**)&dev,10*sizeof(float));
cout << "stat " << stat << endl;
return 0;
}
with nvcc -arch=sm_35 -rdc=true -lcublas -lcublas_device -lcudadevrt -o my simpleCuda.cu compile parameters set I got a warning message
nvlink warning : SM Arch ('sm_35') not found in '/usr/local/cuda-6.5/bin/../targets/x86_64-linux/lib/libcublas_device.a:maxwell_sgemm.asm.o'
nvlink warning : SM Arch ('sm_35') not found in '/usr/local/cuda-6.5/bin/../targets/x86_64-linux/lib/libcublas_device.a:maxwell_sm50_sgemm.o'
`
. In this link I see that it can be ignored. But I don't want to ignore this message. I compiled this simpleCuda.cu with the same compilation parameters set on different computer with cuda-5.5 compilation tool. It does not give me any warning message about architecture linking (-arch=sm_35). I want to get rid of this warning message. These compile parameters are not necessary for this particular code I posted, but further I will need them. I appreciate all your help.

Answer 1

This was apparently a toolchain limitation which was rectified in the CUDA 7 production release.
[This answer was assembled from comments to get the question off the unanswered queue for the CUDA tag]