| spemanig said: Also, can someone please explain to me what SMs are and why the expectation is 2 SMs so I'm not barking like an idiot? |
When Nvidia and AMD build a GPU, they bunch up the GPU cores in units (that has everything those cores need, and call it 1 SM).
One such bunch of cuda cores is called a SM, it stands for Streaming Multiprocessor.
Inside a SM are stuff like: the cores, store/load units, special funktion units ,highspeed memory cache (L1 & shared, and L2), and a interface the connects them all.
this is a diagram from a Nvidia Fermi: https://upload.wikimedia.org/wikipedia/commons/thumb/1/1d/Fermi.svg/368px-Fermi.svg.png
^ each GPU has many of these inside them.
Thats what the 2 or 3, part your in question is about.
Their talking about how many "Cores" inside the Tegra X1 chip, there will be.
The most reasonable assumption is its the standard amount, that the Tegra X1 in all other products comes with.
Why? well its already more powerfull than what they need, if their downclocking it so much.
Why build a bigger chip, if its already more than you need? you dont.
You save a ton of Research & Developement money, just useing the chips Nvidia already has.
