Non-strict cuda driver ? in terms of memory accesses

Accelerated Computing CUDA CUDA Programming and Performance
1

I’d like to know if there is a driver or any technique, such that GPU will not complain during execution even if out-of-bounds memory accesses occur.
For all types of memory accesses (global, shared, and so on…)
Can anyone help ?
Thanks

Or this has nothing to do with a driver ?

2

Install CUDA 2.0; They fixed lot of memory overrun related issues later…

But that said, it is a very bad idea to do all these…Dont use it on production