GPU value vs system value

[porina]

I'm seeing a kinda recurring theme, and the latest round of this nonsense is with the 2080 Super reviews currently going around. What am I salty about? GPU only fps/$ charts. You can't game on a GPU by itself, you have a system around it. How does the system cost affect value? In past LTT videos some kind of system consideration has been given, and it was also used in Scrapyard Wars, so it isn't so bad around here.

 

Of course that then brings us into a much wider question, how much do people spend on the rest of the system? Which number you pick will affect the results. Well, you can chart it.

 

Above is one I quickly knocked up using HardwareCanucks data from their video. The only reason I picked that was that it was the most recent one I saw. I took their stated $ cost for each card, and also their average performance. I can't do fps/$ directly as they don't give that, instead they give a % relative to 5700XT for the average. That's ok too. So I can take the GPU cost and add a fixed "rest of system cost" to get a total system cost, and use that to work out performance/systemcost. I further normalised it so that at each the best value one was at 1.00.

 

So, at zero system cost we're looking at GPU only. The 5700XT takes the lead with the next nearest the 2070 Super about 90% the value, the rest falling further behind. Somewhere around $500 for the system cost, the best value system swings to the 2070 Super, which remains there through to $1400. The 5700XT remains high value throughout, but drops below other nvidia cards as the system cost goes up. The 2080S takes over once you spend $1400 elsewhere. Presumably the 2080Ti would eventually take over, but being realistic, a 2080Ti buyer doesn't figure value high on their priorities.

 

What's a solid upper mid range spec these days? If building today, say, Ryzen 3600, mobo + 16GB ram, case + PSU, garnish with a decent size SSD and maybe HD for bulk. Probably looking 600-700 ball park? And if you were looking at the higher GPUs, you're likely spending more on the rest of the system also...

 

If another site has similar data for more GPUs, I could try repeating this to cover more models. Also I'm debating if it is better to show system cost without GPU as the axis, or add in the GPU cost for total cost as axis. I did it this way as it worked out easier for presenting the numbers on a spreadsheet.

[Fasauceome]

The fps/$ makes sense in a lot of scenarios when you take into account people upgrading GPU only and not building a new system. The rationale behind that is you can factor it in when doing an entire system, as a boutique builder that's sort of my rationale anyways. That's why when reviews said the 2070 was the worst value RTX card out there, I didn't heed that so much because it wasn't relative to total system cost.

[Sauron]

I'm not convinced this is a comparison that makes a lot of sense - yes, compared to the cost of the entire system the relative value of more expensive cards increases but what does this actually tell us? Beyond a certain point ( ~$500-$600 ) the performance gain ( in games ) from spending more on the rest of the system becomes virtually non-existent; additional bling does not significantly impact game performance. By this logic, if I took $1000 and spent it on a gold plated case or even just set fire to it I would be making the 2080ti a better value proposition for me.

 

If you're looking at a 2080ti chances are value is not your first priority - pure performance is, and you can't get that anywhere else. I doubt the relative value compared to the total cost of your system would be a meaningful metric for someone looking to spend that kind of money.

 

It's also worth noting that people generally have a fixed budget and spending more on a gpu means spending less on the rest of the system - which would turn the value calculation into a differential equation, not a linear plot as we see here. There is also an obvious aberration in the plot in that below a certain price the rest of the system would start bottlenecking the card and at some point you wouldn't be able to buy a system at all; here it seems you have been assuming that the cards would perform the same with any system, which is obviously not true.

 

On a base level I agree that some thought should be given to what kind of system should be used for testing, but this kind of analysis is way too simplistic.

[porina]

9 minutes ago, fasauceome said:

The fps/$ makes sense in a lot of scenarios when you take into account people upgrading GPU only and not building a new system. 

I'd argue it still makes sense for upgrading, as the rest of your system doesn't become zero value just because you're not buying it then.

4 minutes ago, Sauron said:

By this logic, if I took $1000 and spent it on a gold plated case or even just set fire to it I would be making the 2080ti a better value proposition for me.

It is. This measure might not be ideal either, but performance / (system cost) makes more sense than performance / (GPU only cost). What's your fps experience for the total amount spent? I could have gone further, I could ague that monitor(s), other accessories, even your table and chair could also count towards the overall value calculation.

4 minutes ago, Sauron said:

It's also worth noting that people generally have a fixed budget and spending more on a gpu means spending less on the rest of the system

This is why I thought a total system cost presentation is better, but it would be more work to display as such.

4 minutes ago, Sauron said:

There is also an obvious aberration in the plot in that below a certain price the rest of the system would start bottlenecking the card and at some point you wouldn't be able to buy a system at all; here it seems you have been assuming that the cards would perform the same with any system, which is obviously not true.

Agreed this is a fault with the data I used and how I calculated. HardwareCanucks used OC'd 9900k to remove as far as possible CPU bottleneck.

4 minutes ago, Sauron said:

On a base level I agree that some thought should be given to what kind of system should be used for testing, but this kind of analysis is way too simplistic.

I don't claim it to be perfect, but it gives a lot more meaningful info than GPU only value.