comparison of year-on-year performance of CPUs vs VideoCards vs Harddrives

[rdekker]

From a scientific point of view, I am interested in whether VideoCard performance increases faster than CPU or Harddrive performance. We like to use it as a demonstration example for a mathematical PC lifetime extension optimisation model, which would give advice whether to upgrade a part of a system or to buy a new one.

In the CPU benchmark menu overview there is a year-on-year performance of CPUs, but not of the VideoCards or Harddrives. As I am at a DataScience / Econometric institute I have quite some expertise available in statistical analysis. Would it be possible to get historical data on performance and date of testing / release? Alternatively, are CAGR (average performance gross rates) known for these groups?

[David (PassMark)]

We have all the raw data, but it isn't trivial to process it into a meaningful graph. It would required couple of days of work.

A mathematical model of when an upgrade might make sense might be interesting academically, but from a practical point of view it might not be so useful.
- Some laptops can't be upgraded at all. Glued shut.
- Laptops that can be upgraded still almost always have fixed CPUs and video cards. (even if RAM and disk can be swapped out).
- All PCs have compatibility considerations. Some examples.
+ You can't put DDR4 RAM in a DDR3 motherboard
+ You can't upgrade a video card if there isn't space in the case
+ You can't upgrade a video card if the PSU doesn't provide enough wattage.
+ You can't upgrade to a NVMe SSD if the right slot is not available on the motherboard.

So many upgrades result in a new motherboard and PSU requirements. If you upgrade the motherboard, then often you also need a new CPU as well as the new MB often has a different CPU socket type. New MB might also force new RAM.
So think your project might be a waste of time unless you also take into account all the practical aspects of doing an upgrade.

[rdekker]

The idea behind the model is that if some parts of a system age faster than other parts, then instead of buying a new system (e.g. PC) it may be more cost effective to upgrade only the fast ageing part. So e.g. upgrade the VideoCard instead of buying a new desktop.This may be better for the environment and needs less capital on short notice:. Of course there may be for each specific case many practical issues, as you indicate, which make an upgrade complex. Yet it is a well applied practice for many industrial systems, like fighter jets, trains. Nevertheless, the aim of a scientific model is to make statements in general. To prove our point quantitatively (in general) we look at the PC market as for that there is a public and well described performance metric (= Passmark scores) with which you more or less can define the "quality" of a part and system. We only need to establish a difference in the "ageing" of the constituing components, because of new components with better scores enter the market. If you are interested I can specify the model in more detail.

[David (PassMark)]

Having a model isn't much good if it doesn't reflect reality. What's the point of the model suggesting you upgrade to a GTX 2060 when it is a desktop part and you can't even open the case of your laptop?

[David (PassMark)]

But we do license out the data if you wanted to do it anyway. See,
https://www.passmark.com/services/market-analysis.php