Which Variables Affect CPU Performance Per Dollar (Value)? Regression Analysis School Project

[afyeung]

If you're reading this I hope you're doing well at home. As the title says, I'm doing a regression analysis (finding a trend using data of variables/factors over time) project for a class and I wanted to do it related to Computers. One thing I've always been interested in is CPU value (performance per dollar) has gotten a lot better and I wanted to observe this trend over time. 

I need your help! 

 

Factors/variables I've been thinking of: Revenue of the PC gaming market, Market caps (company value) of Intel and AMD, Market Caps of internet companies such as Google (biggest consumers of CPUs for data centers), and a dummy variable (either 1 or 0) of if the AMD CPU was the better selling one that year, and also CPU market competition, measured as a ratio of the revenues between Intel and AMD. 

I was also thinking of using LTT video views as a variable because the channel has a big influence on consumers purchasing CPUs. 

 

And for the specific CPU to measure performance per dollar I was thinking of using the i5/r5 equivalent since it's the mid-high end and considered good bang for buck

If anybody has any useful information to help with my project, such as sources for data, factors I could investigate, or any feedback at all that would be highly appreciated. 

Thank you!

[Bombastinator]

Either frames per sec in a given game or flops.

 

A computer is a tool.  Value is based on how a given system does whatever it is that one needs it to

[afyeung]

1 minute ago, Bombastinator said:

Either frames per sec in a given game or flops.

 

A computer is a tool.  Value is based on how a given system does whatever it is that one needs it to

I was thinking of using passmark to measure CPU performance since they have the numbers for tons of CPUs but if you have suggestions on a different benchmark to use that would help!

[Bombastinator]

4 minutes ago, afyeung said:

I was thinking of using passmark to measure CPU performance since they have the numbers for tons of CPUs but if you have suggestions on a different benchmark to use that would help!

The reason there is more than one benchmark is none of them are perfect.   If you’re doing video rendering cinebench could be useful.  If you don’t it isn’t.

[Eigenvektor]

13 minutes ago, afyeung said:

I was thinking of using passmark to measure CPU performance since they have the numbers for tons of CPUs but if you have suggestions on a different benchmark to use that would help!

To compare CPU performance across multiple generations you need a benchmark that has been used consistently over a long period of time with minimal changes, otherwise the results are not comparable. I'd assume passmark is a good start to get a general idea of how performance has evolved. But then again, the benchmark may not take advantage of newer CPU features, which means it may not give you a realistic image of modern day performance that could be achieved if those features are used/needed.

 

I think your general idea is sound. The only potential problem I see is that something like "mid-range" isn't clearly defined and may have shifted over the years. It may be better to take a fixed price (say 200$) and then ask "How much performance did 200$ get me in 20xx?"

 

 

Alternatively you could take both the most expensive and least expensive CPU of each generation, calculate its performance per dollar and then see how these lower and upper bounds have evolved over the years. Or maybe include/use the CPU of each generation that had the best pp$ at that time?

[divito]

As Eigen as pointed out, the technology has evolved many times over, and applications/games still don't always make use of features in chips for several years, if ever. What we do with certain CPUs has also evolved; budget, mid-range, gaming and enterprise has shifted in expectations, along with features. 

 

But as a blunt perspective, I think it's a cool application to see where things have gone and there would be lots of neat things to realize.

[afyeung]

@Eigenvektor I appreciate the response! I think I'll go down the route of taking a fixed price. As for benchmark I think maybe I'll take several such as cinebench and come up with different weights to average out the scores? Really the project doesn't need to be technically perfect its more centered around the analysis and I think even with some flaws I can still get a pretty good idea of which factors influenced the trend the most. 

[Eigenvektor]

2 minutes ago, afyeung said:

As for benchmark I think maybe I'll take several such as cinebench and come up with different weights to average out the scores? Really the project doesn't need to be technically perfect its more centered around the analysis and I think even with some flaws I can still get a pretty good idea of which factors influenced the trend the most.

Yeah, gotcha. As I said, I think Passmark is a good starting point to get an overview of relative CPU performance over the years. If you can find other CPU benchmarks and find a way to weight them to average out their results even better.

 

Just keep in mind that most testers probably used different versions (e.g. of Cinebench) over the years which means the points from one version to another may not be directly comparable. If that's the case, ideally find some tests where the same CPU was tested with both versions. That should allow you to scale the results accordingly with that CPU's performance as the baseline.

 

Sound like an interesting project in any case

[afyeung]

39 minutes ago, Eigenvektor said:

Just keep in mind that most testers probably used different versions (e.g. of Cinebench) over the years which means the points from one version to another may not be directly comparable. If that's the case, ideally find some tests where the same CPU was tested with both versions. That should allow you to scale the results accordingly with that CPU's performance as the baseline.

 

Sound like an interesting project in any case

Thanks for all the help! Good thing a lot of people still test with R15 for cinebench so performance can be directly compared to CPUs since 2013.