05-31-2013, 08:18 PM
When AMD releases their hUMA architecture later this year (which is also likely going to be in the PS4), will oclHashcat take advantage of the uniform memory architecture to jointly use the CPU and GPU to crack hashes?
Forgive my ignorance of all the many hash types that oclHashcat supports, I was just wondering if there were certain hashing algorithms where it might be advantageous to hand off part of the computation to the CPU and then hand it back to the GPU, given hUMA's ability to pass memory reference pointers directly between the CPU and GPU, without having to copy the data between the two.
I know Atom previously has said that he wants to keep the CPU free to do other tasks, but in the case of dedicated cracking machines (especially with vcl) would it be possible to hand off parts of certain hashing algorithms that might run better on a CPU, and then continue cracking on the GPU? Or does every part of every supported hashing algorithm always work better on a GPU?
Forgive my ignorance of all the many hash types that oclHashcat supports, I was just wondering if there were certain hashing algorithms where it might be advantageous to hand off part of the computation to the CPU and then hand it back to the GPU, given hUMA's ability to pass memory reference pointers directly between the CPU and GPU, without having to copy the data between the two.
I know Atom previously has said that he wants to keep the CPU free to do other tasks, but in the case of dedicated cracking machines (especially with vcl) would it be possible to hand off parts of certain hashing algorithms that might run better on a CPU, and then continue cracking on the GPU? Or does every part of every supported hashing algorithm always work better on a GPU?