![]() |
It's that time for a belated look at the yearly trending of recommended system specifications. I'm mostly late because I got side-tracked by other projects, by work, and various other life events. So, apologies for that if anyone was waiting on this data.
As always, the raw data is available here. And, once again, I'd like to pay tribute to Shamus Young, who inspired this series with his Steam data analytics....
Back to Basics...
Just a few short paragraphs to cover the premise of this study and trending, along with caveats and limitations:
As with last year, I have refreshed the Passmark data but the Geekbench data was close enough that I didn't feel it warranted the effort to refresh the CPU data covering 14 years.
The games in this study are picked by myself based on what's big, popular, and/or challenging on the hardware to run. There are a LOT of games released all of the time and the vast majority of them are not difficult to run - i.e. you do not need any sort of capable hardware to play them! So, in reality, there's no reason to even trend that sort of game (though there are a few included due to the stipulations above).
This yearly tracking of data is meant to serve as a rough barometer for what hardware gamers should be thinking of targeting to be able to both play the most games by a certain year, thus helping with decision-making when purchasing hardware in the here and now. The data may also may provide some sort of idea for developers of target hardware profiles for more graphically demanding games.
Ultimately, I'll let you decide what the data means to you.
One big caveat is that I am not correcting for frequency of required hardware in this data. If you have ten games requiring an RX 6700 XT and one game requiring an RX 470, they both get counted once. If you wanted to extract that data for yourself, it's all up there in the spreadsheet. The reason I do this is because I want to counter the bias I have in picking the titles and also even out temporary trends in requirements - as well as over-requirements* (which we know sometimes happens!)
*Sometimes, we see requirements tables and lists with WAY overpowered hardware and struggle to understand why it is there. These results will bias the data towards the higher end and so using only a single instance of each piece of hardware will actually skew things more towards the low end. It's not perfect but it is a balance of sorts...
Saying all of that, it does indicate that my trending is of a conservative nature and hopefully will not be overly biased towards graphically intensive titles while still providing a guide.
This methodology does not extend to memory, where each individual game's requirement is counted.
CPU Trends...
The downward trend for the end-point continues from 2022... |
Due to some software shenanigans, I was unable to exactly replicate the graphs I was previously using so I haven't overlaid the prior year's graph like I did last time. However, the overall trend is clear - a drop from an expected ~14100 points in 2025 to around ~13600 points, continuing from the drop from 2021's expected 2025 result. This still corresponds to a 10600/10700K in terms of single core performance...
Multicore performance also continues to drop from around 7700 points to 7200 points. This time, we're actually shifting down a performance notch from an R7 3700X to a 10600K. This comes as current game titles continue to be unable to really benefit from parallelisation past 6 cores 12 threads.
In fact, when looking at the most common number of cores/threads and their averages across the polled games, we find that most games are requiring a 6C/12T processor, with the average matching the number of physical cores and the number of logical cores staying at 10. This indicates that a good portion of games are still requiring at least an older Intel CPU that lacks hyperthreading, dragging the average down.
Last year, I was predicting that the most required cores and threads would be 6 and 12, respectively (as they are this year). However, it wasn't the case. I predicted it a year early, it seems. However, this year, the prediction is spot on. Let's see if things will increase to 8/16 next year, like I was predicting three years ago!
GPU Trends...
The trend for required GPUs has not really changed from last year. We're looking at a 2080 Super level of performance in 2025 for rasterised games.
But what about ray tracing titles?
Well, I've gone through the data and put together a graph covering that but the issue we have here is that the data is still very sparse....
The main problem we have here is the disconnect between GPU power and memory amount. A 3070/4060 Ti is as powerful as a 2080 Ti but its 8 GB VRAM isn't that great for modern titles.... So, we'll get onto that, now.
Memory Trends...
I am surprised and disappointed that games aren't utilising system memory more effectively. It might not be "smart management" to front load data into RAM but I bet it's a hell of a lot better than requiring data being streamed from storage - especially since we've given up on optane as a storage format...
What is mostly worrying to me is the level of stagnation on show. We're actually regressing to 8/12 GB, like we're approaching an asymptote... At this rate, gaming will no longer be the driving force of system memory upgrades, "AI" will be!*
*Not that I believe in anything "AI" for the consumer at this point in time...
VRAM presents a more logical progression: we're seeing an increase, which is very obviously hamstrung by the choices of Nvidia and AMD, along with the problems of the memory manufacturers getting 4 Gb packages to the market.
I understand the problem - it makes no sense to design and ship a budget GPU with a large memory interface when it needs to be small and cheap to produce and operate (thus reducing board costs).
As a consequence, it makes the memory configuration on the board difficult for the manufacturer because Samsung, Micron, et al. have not increased memory module capacity for years, now.
They're stuck beaten a rock and a hard place.
The percentage of each requirement... |
Looking at the trends for RAM and VRAM in a different format, we see a slightly more nuanced view of things. For system RAM, we are seeing that all game requirements are essentially 16 GB and the other RAM quantities are almost rounding errors. What is concerning to me is that, unlike other periods where a memory hegemon existed (e.g. 2013, and 2015/2017), there is no forward-looking quantity of RAM being requested by game-makers. In both of those times, a larger quantity was also rising in requirements, in parallel. In 2013, 4 GB was being eclipsed by 8 GB. In 2015, we were seeing the stirrings of the rise of 16 GB, whilst 40% of the market was spread across 4 and 6 GB. By 2017, there was a massive jump looking forward to 16 GB with 4 and 6 GB essentially missing in action in the course of a single year.
2023 is an oddball.
We have been through at least two periods since 2020 where memory was SUPER cheap and modern gaming systems really should not be being outfitted with 16 GB of memory. Even now, in a period of relatively expensive RAM availability, 32 GB is only €20-30 more than high-end dual-stick DDR4 16 GB is and has historically been. Why aren't developers asking for it?
I guess, it's partly due to the focus on streaming of data directly from bulk storage (aka SSDs) but that's not going so well*, is it? In a way, I feel like modern game engine design is trying to be too clever for its own good and in the process is making the experience worse for players. Other than those two reasons, I'm drawing a blank, though.
*You might be thinking to point out that PSO stutter is not related to this, but I'm mostly thinking about loading stutter. Saying that, having data already in system memory, pre-compiled and potentially decompressed well-in advance will reduce both... Then we only have to worry about PCIe bandwidth.
So, we're either going to continue to stagnate at 16 GB over the next couple of years or we're going to see another strange jump like we did over the 2015-2016 period.
This chart shows the number of titles which required the indicated amount... |
VRAM requirements show a similar trend. This time, though, the inability of developers to require higher than 8 GB is fully understandable given the paucity of GPUs below the $500 mark with more than 8 GB... Though, the RTX 3060 (presumably mostly the 12 GB variant) is relatively popular, it is vastly outnumbered by any number of 8 GB cards (around 25% of the survey is 8 GB). We're stuck in a rut until GDDR manufacturers actually release 4 Gb modules which will allow the GPU makers to increase RAM on their lower-end, smaller memory bus products (clamshell designs are more expensive to manufacture and design)...
Console Comparisons...
The averaged yearly CPU requirements relative to the requirements in the year of the last hardware release... |
Looking over at console land, we can see that required CPU performance on the PC is not very different from the power of the Xbox Series and PS5 consoles (at the end of 2023, the averaged requirements are 98% of the theoretical performance of the XSX). Four years into the current console generation, we're looking at games not pushing CPU performance beyond what the PS5/XSX can do - with engine technology generally focussing on GPU processing.
Out of the last three console hardware changes, we observe the lowest level of CPU advancement over the same time period. Both the PS4/XBO and Pro/XBOX hardware saw around +50% for single-threaded performance in the third year after release. We're currently at +8%.
You could argue that this is all the CPU performance that we actually need. That may be the reason - and that everything else is just pure optimisation of existing codebases to improve hardware utilisation, reduce bottlenecks, and manage data more efficiently... Let's hope that happens!
The averaged yearly GPU requirements relative to the requirements in the year of the last hardware release... |
GPU game requirement advancement tells a similar story: GPU "power" is powerful enough to make almost any game. We're looking at the power of an RX 6600 (non-XT) here in 2023, which is around 80% the estimated performance of the GPU in the XSX in rasterisation workloads.
At this point, it seems to be clear that VRAM is a bigger limiting factor when not-taking ray tracing performance into account.
Wrapping up...
This console generation continues to refuse to take-off with generationally-exclusive titles still feeling few and far between. Hellblade 2 just released and is a graphically demanding powerhouse but which scales well. Alan Wake 2 also pushes the boundaries of graphics but other than those, we're not really seeing very demanding titles.
In PC land, I think that we are in a good situation with regards to CPU and platform performance. Getting an AMD 3D V-cache Zen 3 part or a cheap Zen 4 part will provide the user with more than sufficient performance to run games for the next few years. RAM and storage are pretty cheap (despite increasing in price over the last half a year).
Unfortunately, both motherboards and GPUs spoil the party, price-wise. In a sense, this is understandable. Motherboards (at least on AMD's side) have to account for decreased sales due to the longevity of AMD's platforms - thus, a higher buy-in price is logical as it is ameliorated over a longer period of time. GPUs fall flat mostly due to the cost to achieve good ray tracing performance (> $500) and because VRAM is such an issue. However, as I said, this appears to mostly be a problem due to the stagnation from the memory module manufacturers.
This year, I attempted to try trending ray tracing requirements. However, it wasn't possible - there are too few titles to trend that include RT and even fewer which actually have separate RT recommended requirements. Maybe I will be able to start when reviewing 2024 games.
Until then...
1 comment:
Yeah, PC CPU requirements have essentially stagnated, for which I'd place the blame wholly on MS/Sony. MS for relegating this gen to no mid-gen upgrade, and Sony (PS5 pro) supposedly doing a mid-gen release but with no CPU improvement.
Post a Comment