Page migration vs direct cache transfer

[Yechen Liu]

Hi there,

I was confused about how unified virtual memory is achieved.

If there's a GPU L1 cache miss, the cache block will be directly brought into GPU L1 cache. Does it mean the GPU page fault will never happen? Could you please provide an example that will cause page migration rather than cache to cache transfer.

Thanks,

Yechen