Introduction Flashcards
What is computer architecture?
Designing a computer that is fit for its purpose
Moore’s Law
Every 18-24 months, the number of transistors we can put on a chip doubles.
What do computer architect’s try to do in response to Moore’s Law?
Double processor speed, energy efficiency, and memory every 18-24 months.
Moore’s Law is about what tech we can build, architecture is about what we can do with this tech.
What is the “memory wall”?
Processor speed and memory capacity have been doubling every two years, but memory latency only improves by 1.1x every two years. The more time goes by, the larger the gap between processor and memory access speed.
We use caches to mitigate the problem.
What are the two kinds of power consumtion?
Dynamic power: consumed by processing activity.
Static power: consumed when a processor is ON but idle.
Active power equation
P = 1/2 * C * V^2 * f * alpha, where
C = capacitance, roughly proportional to chip area
V = power supply voltage
f = clock frequency (GHz)
alpha = activity factor, the proportion of transistors that are active.
Static power and leakage
A transistor is like a faucet that is controlled by voltage. The greater the voltage, the more tightly closed the faucet is. If voltage is high, you get a lot of leakage.
So as voltage increases, dynamic power increases. But as voltage decreases, LEAKAGE power increases. So there is a spot in the middle where total power is at its lowest.
Fabrication Yield
Chips are manufactured on wafers that are then cut into chips. Each wafer has some number of defects. Large chips means that a larger proportion of the wafer is wasted, smaller chips means a smaller proportion.
Smaller chips means both more chips and better YIELD.
Yield
Num of working chips / num of total chips