Faster CPUs solve the same computational problems in shorter time frames. The same applies to faster GPUs.
Therefore, replacing with faster components can reduce the computer's power consumption.
On the other hand do some components need more electrical power for instance for ventilation.
How do you estimate which components to replace by which, find the bottlenecks, to not only save electric energy but also improve the computer's performance?
Replacing components in an existing computer with different ones is often hard to impossible.
CPUs usually don't have fixed performance at fixed use of electricity, but there is a curve where you change the voltage of the CPU, and it increases both performance (linear) and power usage (quadratic). The limits are at the low end where the CPU stops working, and at the high end where the CPU self destroys through heat. If you compare an M1 and M2 processor which are practically identical, M2 can produce better performance at higher power usage, slightly better performance at same power usage, same performance at lower power usage (improving battery life), and lower performance at much lower power usage. Chosen at runtime.
Big savers are turning brightness of the screen down, smaller screens, and replacing spinning hard drives with SSD drives (at much higher cost, and AFAIK higher energy cost to build the drives).
It's really hard. As a first approximation, my sense is that you basically can't, in any reliable, scientific way, in most cases. Modern computers are complex, and it's hard for an average consumer to predict exactly what the impact on performance or energy usage will be of any particular change to one component. In general, if you want to look at how you reduce memory consumption, you can start by looking at what laptops do; they are engineered to trade off performance for lower energy consumption.
Of course if cost is no object then you can just replace everything. But in practice cost is always a consideration, and it's hard to predict exactly how much benefit you'll get from any particular change.
Let me give you some ideas of some of the challenges. First, there are many factors that affect overall performance and memory consumption. Let me give you a simplified example. Suppose that:
Now suppose you are considering replacing your RAM with the newest, latest technology that is 2x faster and uses 2x less energy. What will the impact be on these tasks?
See Amdahl's law and Gustafson's law.
In practice, most tasks look more like task V and depend on many factors, so replacing any one component is likely to yield only small benefits on most tasks, because those tasks rapidly become bottlenecked by the other components you haven't replaced. A few tasks will see a big benefit and some tasks will see no benefit. Moreover, because workload varies from person to person (you might run a different mix of applications than I do), experience from one person doesn't necessarily translate to others.
To predict the impact on performance or energy, you need to be able to measure all of the factors that affect a particular task's performance and how they affect it (this is usually hard), measure the effect on those factors of replacing a single component (this may be hard, or may be feasible using public benchmarks; it depends), measure the workload and mix of tasks (this is tedious at best), and then simulate all of those. This is beyond the capabilities of most consumers, and is very challenging even for those who are deeply experienced in the field.
If you look online, you'll find a lot of heuristic advice, opinions, and personal experiences. Some of them may be valuable. You may find statements that replacing a HDD with a SDD will really speed up boot times; if you have 8GB of RAM or less, memory is cheap and upgrading to 32GB might give you a big boost because some tasks might have been thrashing (e.g., a browser with lots of tabs open); that replacing your CPU with a mobile CPU or underclocking it can save a lot of energy; that buying the highest-end GPU you can afford will best improve performance of games, but will cost more in energy; and so on. You may also find advice that, if you're an average consumer, it's often best to (wait a few years and) upgrade your entire computer to a new computer, rather than replacing a single component, because a new computer will have new components that are better in every respect and in a balanced way, whereas replacing a single component may just leave you bottlenecked on the other components, reducing the benefit. Those are fine as far as they go, but they fall a bit short of a quantitative prediction of exactly how much performance and energy consumption will improve from a specific replacement on your particular computer.
