SM: Did you get started because of pressure from the VC’s or did you feel the market was turning? AA: I thought the timing was right in 2004, so we formed the company in October of that year. What is interesting is when I go back and look at the time of our VCs presentations, and even in 2005, multicore still had not gotten hot even though we had built a tiled multicore processor in 2002. Our VC presentation did not have the word multicore anywhere in it – we called it all kinds of things, but we had not figured out what this new category was going to be called. We called it Tiled Processors, and so on, but the word Multicore only got popular in late 2005.
SM: Who came up with the terminology? AA: I didn’t. The word multicore was used by others first.
SM: What dimension of the architecture was hitting the wall in single core processors? AA: There were three things.
First, in the past 30 years, processors kept getting more and more transistors and became more and more complex. The clock frequency kept increasing. After 2002, you simply could not take it any further. All the single processor mechanisms we knew about then were giving diminishing returns. Everything had gotten so big, the pipeline was full and there were no more benefits.
A second cause was that we had hit the power wall.
Third, wire delays became a big issue, in that, the speed of transistors themselves was not the big issue, rather the time it took for a signal to propagate on a wire became the issue. What that meant is, building bigger and bigger structures was not possible anymore. The bigger the structure, the more power it would require and the more latency placed on the wires.
Thus, the three causes were all interrelated, and there was no way to progress further in a single core design.
SM: How do you differentiate between multi-processor and multi-core? AA: Multicore is when you put multiple processors on a single chip.