Over the few days I studied computer architecture and it really changed my view of Artificial Intelligence.
I started with RISC-V International. Learned how processors understand instructions and run programs. What I find interesting about RISC-V is that it is open-source so researchers and startups can make their own hardware for Artificial Intelligence work.
Then I looked at Verilog and digital logic, which showed me how hardware circuits are designed. It was cool to see that Artificial Intelligence accelerators and tensor processors are made from these concepts.
I also learned about simulators like gem5, where researchers test and improve architectures on computers before they even make the hardware. That part really surprised me.
The topic that excited me the most was Artificial Intelligence accelerators like Tensor Processing Units and Neural Processing Units. Modern Artificial Intelligence models need a lot of computing power and normal Central Processing Units are not enough anymore. Special hardware is becoming very important for Artificial Intelligence.
My biggest takeaway is that Artificial Intelligence is not about software anymore. The future is about Artificial Intelligence hardware and software working together as one system. From the compilers to the memory systems, to the accelerators everything is connected now.. That is where the next big innovations will happen.