Hi,
This may be a known where, in compatibility profiles where the indirect commands are located in system memory, the base instance is not propagated to the GPU. If this is the case, please try placing the indirect commands into a buffer object. There is no real advantage to using system memory for indirect commands as the operation will be faster if the commands are already located in GPU memory. In any case, AMD recommends using only the core profile of OpenGL where possible, under which this behavior would be illegal.
If using a buffer object does not resolve the issue (or if a buffer object is already in use), please let us know.
Thanks,
Graham