OptiX 3.8.0 Compilation problems with MS Visual Studio 2008

We have to build our application with VS 2008, which is also listed as supported Compiler for OptiX 3.8.0.
But we are facing compilation errors when building the OptiX Samples.
When building the sutil lib which is used also in our project code.
Error message:

OptiXMeshImpl.cpp
1>MeshBase.cpp
1>…..\sutil\MeshBase.cpp(862) : error C2039: ‘at’ : is not a member of ‘std::map<_Kty,_Ty>’
1> with
1> [
1> _Kty=std::string,
1> _Ty=int
1> ]
1>…..\sutil\OptiXMeshImpl.cpp(157) : error C2039: ‘at’ : is not a member of ‘std::map<_Kty,_Ty>’
1> with
1> [
1> _Kty=std::string,
1> _Ty=OptiXMeshClasses::GroupGeometryInfo
1> ]

It seems that a later version of std::map is used as it is available in VS 2008.
Are there any recommondation how to fix this?

You could replace the code by iterators and .find(key) instead of .at(key).
This should do the same. (Untested, that part of the code isn’t used inside newer OptiX versions anymore.)

// curr_material_number = m_material_numbers_by_name.at(curr_material_name);
  MeshMaterialNumbersMap::const_iterator it = m_material_numbers_by_name.find(curr_material_name);
  if (it != m_material_numbers_by_name.end())
  {
    curr_material_number = it->second;
  }

Other than that, sutil is just a number of utility functions for the OptiX SDK examples. There is no requirement to use any of that code in own applications.

The list of compilers we use is limited to the ones which are supported by the CUDA toolkit version with which a particular OptiX version was compiled. At this time the recommended combination under Windows would be OptiX 3.9.1, CUDA Toolkit 7.5, and MSVS 2013.
You should strive to keep up with the most recent OptiX and CUDA versions for bugfixes, performance improvements, and current GPU support.

Thanks.
With this hint we can work further.
Unfortunately we have to use VS2008 because this code is developed
for CatiaV5 R24 and Dassault forces us to use VS2008.

Just for those compiler problems we normally avoid using such classes like std::map
which may not work properly on every compiler or version.
Instead we use for our software own classes.