CUDA Managed Memory inside an Enclave

[SANJAY_S_]

Hello !

I need to offload certain parts of my application to the GPU and looking at using CUDA managed memory to reduce the hassle of maintaining 2 buffers.

Would it be possible to access such memory from within the enclave, and if so, what can be the extra attack vectors ?

And a more dumb question : Can managed memory reside inside the enclave ? ( I suppose not because cudaManagedMalloc would entail a system call )

Thanks,

Sanjay

[Anusha_K_Intel]

Hi,

Intel SGX does not support CUDA.