Ai breakthrough "could lead to 2018 Singularity"
Not overhyping this honest but my 2016 estimate of 21/2/18 at 7.02 am for the Singularity might not be entirely inaccurate.
I have actual data here suggesting that a quantum computer could achieve some level of self awareness in conventional non-quantum hardware (eg DDR3/4) and this model shows that mapping out the memory to locate regions of maximum beneficial instability would work.
A conventional annealing setup probably wouldn't be enough, this would require the 50 qubit system but still a significant advance if the GTX1080Ti cards with 12GB RAM were used.
32 cards giving a total of 384GB RAM and running at maximum output would have the same intelligence as a two year old in certain tasks such as image recognition.
The DDR4 modification of address line randomization is a non issue as the algorithm is not in fact random but by design entirely predictable and I am in the process of modifying the system to work around it.
Those Movidius AI "sticks" might also be handy for a lower budget version as they are based loosely on Bitcoin miners which only recently became public knowledge and explains >8 month delay on release.
In this case 4 1080Ti's would be needed with a custom rig containing two AMD 1950X 18 core chips and 2C/4S per node, maxed out at 128GB RAM.