Build of the Week: Feb 16 and 23, 2025
- Tomiax
- Feb 23
- 20 min read
What will the future hold? Well, let me get my crystal ball. Oh, when I look through it all I see is a rotated upside down and distorted image of the environment on the other side of it. What about my magic 8-Ball? "The future is uncertain." Maybe there is wisdom in that plastic sphere. It's not that we can't make predictions about the future. We do all the time. We can often look at past and present events, and then extrapolate forwards. Sometimes this process is highly intuitive. You caught that ball by predicting where it was going to be, after all, didn't you? Sometimes, especially in systems governed by physical laws, our predictions might even be both precise and accurate. If we know the amount of gun powder in a cannon, the size of the ball in it, the orientation of the cannon barrel, and the windspeed that day, then we might have a really good idea of where the cannon ball will land once fired. However, no matter how much we know, there is always the possibility that something unpredictable will happen, such as a change in the winds. But this all begs the question:
Why are we trying to predict the future, anyways?
Because we want our gaming system that is both expensive and takes time to build to be useful for more than just a couple of years before needing to be replaced, that's why. If we wanted to replace our whole system every couple of years, we'd save money and be dirty console gamers! PC Master Race!
LOL
So if we can't perfectly predict hardware demands of future games, then what's the best we can do? The absolute best would be to look at the entire history of the most demanding games from each year, break down the system requirements of each of those games into measurable specifications, and then produce a best fit of that data to a function that can be used to extrapolate each statistic into the future. Two things:
We aren't going to look at the entire history of PC gaming on a Sunday Post
We aren't going to take the time to find the most accurate mathematical models of every data set.
So, what are we going to do? We're going to do the next best thing. We're going to gather as extensive a data set as we have the motivation to produce, and then try to notice trends in the statistics that we can easily extrapolate forward. These won't produce the most precise predictions. In fact, we're not going far enough into the statistical analysis to provide any meaningful margin of error in the end results. However, that won't really matter anyways since there are multiple factors that can change the rates at which hardware demands do and don't change over time. The natural fluctuations in hardware demands that result from new game development techniques and escalating consumer expectations will hurt the potential accuracy of our models even if we did take the time to produce precise mathematical models of the history. So, we're going to loosely look for trends that are easy to extrapolate forwards, but what data set are we going to be using for this?
How we have chosen our data set is quite arbitrary. The broader the data set we have, the more we could account for the variance produced by unforeseen factors as part of our model - and therefore the more valuable taking the time to produce such a model would have. However, we're not spending the time to put in hundreds of data points for each specification. Instead, we're going to pick about 20 examples for examination, one for each year from 2010 to 2024. For consistency, we should try to pick examples of titles that represent the same level of demand relative to other titles of the same year. We're picking titles at the higher end, and we're using the higher "recommended" system configuration whenever available instead of just the basic "required", for three reasons:
It's easier to be consistent in our selection criteria.
What ever games we're playing, we want to actually enjoy them - not just barely "run" them.
If our predictions wind up underestimating the system demands of the future, we'll have more margin for error if we overshoot our estimates.
Once we see the data, we're going to use it to predict what top tier gaming demands might be like ten years from now in the distant year of 2035.
So, here is a table that summarizes the data we have collected today:
Keep in mind that the prices listed for components are based on availability found at the time of writing the post. Prices may vary, and are based on Canadian Currency in a Canadian market. The admin of this website might earn a commission whenever links that lead to Amazon are used to make purchases, but these links do not affect the final price to the consumer. Furthermore, we will still always link to the lowest price we found. Remember to add taxes and shipping to these prices when budgeting, and consider cost-saving strategies like waiting for sale prices or finding used parts.
Analysis
CPU Cores and Threads
A comment that we will make and refer back to a few times as we look at the data is that the year 2017 seems to be a boundary year for CPU demand in games. Before this year, there seems to be a fair bit of inconsistent fluctuation. This is likely due to a tug 'o war between needing to put more CPU utilizing features into games, and then finding ways to better optimize the programming that supports those features. The year 2017 itself is a random low point on CPU demand. Maybe there just weren't so many games that year that put a lot of load on the CPU? From 2018 onwards, however, the increase in demand on CPU seems to be much more stable. Unfortunately this doesn't give us a lot of data to work with, but it seems that the required number of cores seems to increase by two units every three or so years. Extrapolating this forwards, we're going to need about 16 cores, maybe 20, for a top-end gaming system in 2035.
When it comes to threads, the standard has been two threads per core since about 2011. It's entirely possible that this could change in the future, but that's hard to predict. The thread to core count was exclusively 1:1 for the first few decades of the existence of CPUs. It's been just over two decades (since intel introduced hyperthreading in 2002) that CPUs with more than one thread per core have existed. Since then, the standard of two threads per core has gradually become more the standard, and it's been at that standard now for a long time. So, unless something specific happens that makes triple threading more efficient, it's more likely that CPUs will just continue to have more cores with higher clock speeds. So, we'll probably want a CPU with twice as many threads as cores. So, 32 to 40 threads is the target.
CPU Clock Speeds
The year 2018 onwards has had an average progression of increased demand on CPU clock speed of 0.1 or 0.2 GHz per year. By 2035, that puts us requiring at least 5.0 GHz, or 5.4 GHz more likely.
RAM Capacity
According to our data, the required RAM capacity seems to double every 4 to 8 years. If it doubles every 8 years, then it will double once between now and 2035 landing us on 64 GB. However, we suspect that an assortment of factors will require RAM capacity to escalate at the faster rate, meaning the required capacity could double a second time up to 128 GB. Some of the factors we are looking at that make us think this include:
Larger and more detailed environments require more RAM to load and track:
Zone boundaries.
Terrain features.
Object locations, motions, and properties.
Current whereabouts and actions of both player and computer controlled characters.
More accurate and precise physics engines in games will require more RAM to store the data and formulae required to run complex computations.
More nuanced AI decision making for characters controlled by the computer requires extra RAM space to store all the rules and relations via which the computer will make decisions.
Multiplayer online games require the loading of a lot of data to track the character information and game states of each player.
RAM Clock Speed
The required clock speed of RAM seems to increase a lot more steadily than any other specification we've covered so far. Sure, it will jump and then stay steady for some number of years, but if we round down to 1200 MHz in 2010 and average a 200 MHz increase every year since then, we wind up matching the starting and ending points very closely while not straying too far from the in between points in the process. If we extrapolate out the same rate of increase for another ten years, we predict needing RAM with 6000 MHz clock speed in 2035.
GPU CUDA Cores
Every once in a while we get to learn something new and, for me, this is one of those moments: When researching the needed core counts for GPUs over the years, I learned that the architecture for a Central Processing Unit and a Graphics Processing Unit are radically different. CPUs are meant to handle any computational task that comes its way. GPUs are much more specialized, and the bandwidth of their specialization vary from one architecture to another. We inadvertently predicated this section largely on NVIDIA's GeForce GPUs by tracking required CUDA cores before realizing that CUDA cores are proprietary to NVIDIA specifically, and the Stream cores used in AMD's graphics cards aren't exactly equivalent in all tasks. Although the performance of AMD's Stream cores fall behind CUDA cores in applications like machine learning and modeling scientific simulations, they do perform roughly equally for graphics rendering. Since graphics rendering is what we are mostly concerned with when it comes to gaming, we could treat the core count as roughly one to one to a good enough approximation. However, NVIDIA's current graphics cards do generally have one edge over AMD's: Raytracing technology in NVIDIA's recent GeForce RTX cards is much more mature and well developed than AMD's. Since most game developers are leaning progressively harder into ray tracing, even with talks about it not being able to be turned off in future games, this will encourage us to stick with NVIDIA for now. This is why I never corrected my focus on CUDA cores to a more balanced approach when I realized what I was doing: I'm going to stick with that architecture for now anyways.
Now that we've explained the unintentional bias in focusing on CUDA core stats, what can we learn from them? Well, the required amount of CUDA cores definitely trends from lower to higher, but the pattern it follows is not consistent at all. At times, the count goes against the overall trend. We will see similar patterns for VRAM and GPU clock speed when we study those. Perhaps this has a lot to do with games becoming better optimized to make use of architectures that exist as they are developed. However, optimization can only go so far before we are as efficient with a system as we can be. At that point, the only way to get better performance is to have more raw power to work with. That means more CUDA cores, more VRAM, and more GPU clock speed. In terms of projected CUDA core counts required ten years from now, we simply assume that another 1000 will be needed every other year to keep up. We don't have enough consistent data to see any real trend, but following that assumption we "predict" - and "predict" is a kind word for it - that we will need 13000 CUDA cores for top end gaming in the year 2035.
GPU VRAM
The demands on VRAM in the GPU will likely just continue to go up, and some of the reasons why will overlap with why regular RAM demands will continue to increase. Larger and more detailed environments with more objects in them will require more VRAM to load models, textures, and governing rules. The increasing demands on VRAM seem slightly more predictable than CUDA cores. It seems that if we increase the VRAM by 2 GB every other year starting in 2012, then we wind up with a model that matches the data well enough. Extrapolating that forward to 2035 means we are going to want 26 to 28 GB by then.
GPU Core Clock Speed
The GPU Core Clock Speed demand seems to have a jump of 200 to 500 MHz every three to five years. So, in a ten year span, that amounts to about a 1000 MHz increase. 1700 MHz plus 1000 MHz lands us at 2700 MHz.
Storage Capacity
Games today require more storage than they did 15 years ago. There are multiple factors that influence how quickly storage capacity demands increase for games.
Storage devices that have more capacity at lower prices give more room for game developers to put more into their games.
Faster storage devices allow game developers to load more data all at once without hurting loading times.
Players expect larger and more detailed environments, and more detailed models and skins, all requiring more storage.
Improvements to online servers can sometimes suspend the progression towards higher storage demands as they can host a lot of the environment data like maps and interactive objects.
Developing better coding efficiency reduces the need for more storage to achieve the same things.
We can kind of see this tug 'o war on our table. Every once in a while the required storage capacity will double or triple, but then it will also often remain steady for a few consecutive years or even slip backwards a bit. If we assume a similar trend, then we might expect required storage to double twice, stagnate a fair bit between doublings, and maybe slip back a little at some point. So, we land on a figure of about 700 GB storage being required for the most demanding games in 2035.
Selecting Parts
We are trying to design a system that won't need to be overhauled every few years to remain top-tier for gaming. So, we're going to be looking at our predicted required specs, and attempting to match or beat them whenever possible. That said, trying to beat these predicted standards is expensive, and might not be possible for all components. So, we are going to be prioritizing certain components and maybe compromising a little on others. The RAM, for example, fits into slots that are easy to access when the side panel of the PC case is opened. Conversely, the motherboard is not so easy swap out later. Since everything is connected to the motherboard, updating that foundation later would require us to completely break down and reassemble the whole system. This means we also really don't want to find ourselves needing a CPU upgrade later. The socket on motherboards that are required by a CPU tends to change every so many generations. This means that updating a CPU later leads to a strong possibility of also needing to update a motherboard, which we just finished explaining we don't want to do. Therefore, we'll be starting with the CPU and Motherboard combo, and we absolutely will not compromise on these components.
Central Processing Unit
In our prediction, we think there is a good chance that we will need a CPU with 16 to 20 cores, and 32 to 40 threads, and a core clock speed of 5.0 to 5.4 GHz. As stated in our prior paragraph, this is not a component we can compromise on if we really wish to avoid with all our might a total system rebuild in less than ten years. Therefore, we're suggesting a more expensive CPU than we normally might:
24 Cores exceeds our projected needs.
32 Threads meets our minimal projected needs.
Boosts to 6.0 GHz clock speed which exceeds our projected needs.
Has integrated graphics (foreshadowing).
Priced today: $639.99 @ newegg
Addon about the CPU: We originally were going to save $40 by selecting the very similar i9-14900KF, witch does not have integrated graphics (notice the "F"), but we later came back and decided to instead use the i9-14900K (notice the lack of the "F") which does have integrated graphics (foreshadowing).
Motherboard
Again, this is another component that can't be compromised on if our goal is to avoid complete system reworks. We need a motherboard that will support the CPU we picked out, but we also need a motherboard that will support all the other components we choose today while also supporting any upgrades we might predict needing to make in the future. So, we can specify that it has to support DDR5 RAM up to (and possibly in excess of) 128 GB. We also want to make sure that the motherboard supports RAM speeds in excess of our 6000MHz prediction since this could really hurt the future performance of our system if we underestimated how fast the RAM will need to be. We also need to consider what storage we can connect to it. Depending on how large future game files might get, we care about not just how large our storage capacity is, but also how fast it can communicate with the rest of the system. So, we need our motherboard to be compatible with the best standards for storage data transfer rates we can get, and you currently can't beat M.2 5.0x4. We might not necessarily cap out this feature with the storage we choose today, but we'll cover that when we go over selecting our storage. The point is that our motherboard has to be able to support it in case we need it. This motherboard fits the bill on all these fronts, and adds decent good internet connectivity by today's standards, all while looking great for our expensive art-piece system:
Has the LGA 1700 CPU socket needed for our chosen CPU.
Supports up to 192 GB of DDR5 8000 MHz RAM, which exceeds the capacity and speed requirements we project needing.
Has 5 slots for M.2 storage devices, two of them up to the 22110 size standard.
Supports M.2 5.0x4 storage devices.
1x 2.5 Gb/s ethernet and Wi-Fi 7 for network and internet connection.
Priced today: $459.99 @ Amazon
CPU Cooler
This is the first time we have to make a sincerely tough choice. The CPU does not include a cooler out of the box, and the odds that we'll want to stay with a basic cooler for the whole of ten or more years is very low. The reason for this is because, if we need this CPU to keep up with our projected demands, we will have to rely on it's boost clock speeds. We also might look into overclocking in the future if future demands on the CPU exceeds our expectations. Both using the boost clock and overclocking draws higher power and produces more heat. So, we will need a robust cooling solution to meet this demand. However, the CPU and motherboard alone are already quite expensive. Now, weigh that against the fact that changing the CPU cooler sometimes requires taking the motherboard out of the case to swap out a mounting bracket, which then reopens the conversation about how we don't want to tear down and rebuild the whole system during this time. Further to that, there is also the short term vs long term cost analysis. With cooling, and knowing that we'll need something very robust eventually, we want to compare the cost of something good enough right now to something that could be good enough for the whole ten years. The first thing we might look at is a top-end air cooler, which is an affordable way to go, but any air cooler will likely struggle to maintain consistent thermals when pushing a CPU like the one we picked to its limits. So, we're going to get a very robust water cooler with a large 420mm radiator, but we're not going to get the most expensive one we can find. We're getting something that shouldn't struggle keeping our CPU cool under loads, and that should help the water cooler last.
Has a 420mm radiator, which is one of the largest form factors available and enables a lot of heat transfer from liquid to air through fins with minimal fan speeds.
Lots of cooling power.
Lower fan speeds puts less stress on fan bearings to preserve longevity.
Keeping the water in the system cool efficiently through a large radiator also means less need for the pump to run fast.
Preserves the water pump for emphasizing longevity.
Priced today: $209.99 @ memoryexpress
Random Access Memory
RAM is very easy to swap out in a system and, although we don't want to front the costs of changing RAM every couple years, the cost for performance and capacity of RAM does tend to reduce over time. The other observation is that the cost of RAM available now scales more aggressively with capacity than it does speed. So, we could spend about $100 on something that's at our 10-year projected 6000 MHz speed demand, but that barely fits our current 16 GB capacity demands. We could spend less than $150 to keep the same speed and double the capacity to 32 GB, and probably get 3 to 5 years before we have to start keeping an eye on the demands of top-tier games. If we wanted to double that again to 64 GB, keep the same speed, and potentially still need to replace it for a 128 GB set, then we are likely spending close to $400. If we want a 128 GB set... Well we didn't even find any kits available at the same 6000 MHz speed standard. Sigh, I think we can save money here and go with a 32 GB set of fast RAM. There's a small chance best case scenario that we might not need to replace it in less than ten years if diminishing returns results in less than projected demands on RAM capacity. Conversely, it's also possible that our projection could err on the side of underestimating future demands on RAM. So, we could blow a bunch of money on the biggest, baddest set of RAM we can find and then still finding ourselves needing to dump a similar sum of money again later to replace it. If we go the suggested route of spending less money now and need to upgrade the RAM later, then at least we didn't spend as much money up front. So, let's get a 32 GB set with good speed, and pick something we like the looks of. We suggest something like:
32 GB is short of our projected need of 64 to 128 GB:
If our projection overshoots future demands, we might still get 10+ years with a 32 GB set.
If our projection is accurate and we wanted to just get the best RAM now that wouldn't need to be replaced, then we would have to prepare ourselves to spend $600 or more on a slower RAM set since we didn't find anything today that has a 128 GB capacity with 6000+ MHz clock speed.
DDR-6400 is faster than we project needing, but our projection is conservative in this case.
Priced today: $124.99 @ Canada Computers
Storage
A 2.5" SATA 6.0 Gb/s form factor SSD is easy to add any time that we want. Even M.2 drives, although often requiring the removal of the graphics card and/or heat plates on the motherboard to access their ports making them tougher to add later than the 2.5" variant of SSD, can be added later without requiring rebuilding the entire system ground up. (That said, we might not need to worry about working around a GPU before upgrading storage (foreshadowing).) So, we're not too worried about having tons of storage capacity for things like files and photos. We care more about having a fast storage device with enough capacity to host our operating system and a good handful of our current favourite games. Then, when we need more storage later, we can add it while still making use of the originally installed device. We can easily take the same attitude with the speed of the storage device. We can put in a M.2 PCIe 4.0x4 device that's more than fast enough for operating systems and quickly loading our current games. Then, so long as we are smart enough to keep the ports that support M.2 PCIe 5.0x4 available, we can make use of faster storage devices when we need more capacity later. By then, the technology behind M.2 PCIe 5.0x4 will likely be more available and more affordable. So:
2 TB is plenty of room for our operating system and a handful of our current games.
Read and write speeds of 7400 Mb/s and 6400 Mb/s respectfully is more than fast enough for initial boot of the operating system today's gaming standards.
Priced today: $165.99 @ Amazon
Video Card
The choice for which graphics card we will go with is another point that's really difficult to make. Instinct tells us that we want to stay top-notch to match some of the other components we have selected so far, but the truth is that right now the top notch 40- and 50- series NVIDIA GeForce graphics cards are either unavailable or costing more than some cars. Meanwhile, the latest AMD Radeon graphics cards have not hit the market yet. A brief scan of the GPU market today almost makes us want to skip getting one altogether. Instead, we can try to salvage one from a previous system (such as the GeForce 1070 I still have in my personal system), or try to find something second hand. Or, we might just skip the damned thing altogether if the games and work we are doing right now aren't too demanding for graphics rendering. (Did I foreshadow this enough?) That said, if we must pick something, then I guess we'll go with this and save money for a better one later:
Eh, AI says it's still fine for most modern games and it doesn't cost $2,000?
We plan to replace as soon as GPU prices are reasonable again, if ever.
Maybe intel will save us all by making more of a splash on the GPU scene?
Priced today: $301.95 @ shoprbc
Power Supply
Power supplies don't tend to be as expensive as some other components (graphics cards, am I right?). So, we can usually find something pretty good for not too hurtful of a price. Not to mention that wiring one in is a pain. So, we're going to give the power supply the same treatment we did the CPU cooler: It's not a priority to get the most expensive thing on the market, but we do want to make sure that it does more than enough for our future uses. Current estimated wattage demand of the system is about 511W, but that can easily jump to 750W or more if we go ahead with our intended upgrades (larger RAM set, more storage devices, newer and better GPU). Usually we want our estimated wattage demand to be between 50% and 75% of the power supply rating for a few reasons. First, most power supplies run at maximum efficiency somewhere in this range. Second, if we upgrade to more power hungry parts than anticipated, we'll still have room to grow. Finally, a power supply that is operating at 50% will last longer than one being pushed to 100% before beginning to depreciate, and one that only has to put out 50% will still work after it has begun to slow down. We also want something fully modular. This will enable us to choose, and change out later, only the cables that we actually need to make cable management a little bit easier. The last couple of things we demand from out power supply are these: It has to have a decent efficiency rating. It has to be highly rated and reliable. So, we have this:
Overkill for the 511W we are designing today, about right for the 750W that we might jump to when we do future upgrades.
Fully modular for cable management and making adjustments later.
80+ Gold efficiency rating.
Priced today: $306.99 @ PC Canada
Case
The required budget for our PC before we even look at cases is just over $2200 (before taxes) so far. So, what we do with the case will depend on how much we have to spend. This post so far has not assumed that the budget is super tight since we are trying to design future-proof system, and we won't start now! This is a build we plan to have, and therefore look at, for a decade or more. We want a case that will offer room to build and upgrade, great airflow options, front USB connectivity options we want, and sleek aesthetics. I think we've chosen this in previous weeks, but it has all we're looking for:
Great cable management design.
Very clean design and angles.
Roomy interior.
Comes with fans.
Priced today: $379.99 @ BestBuy
Summary
Parts List
Costs about $2600 plus taxes.
Upgrades
The CPU and motherboard combination should be good for the ten plus years we planned for, and so should the water cooler we selected. There is a reasonably good chance that we'll need to upgrade our RAM capacity, but; the RAM we selected should be very good for a while before we have to change it, it's easy to swap out the RAM without disrupting anything else in our system, right now getting a larger RAM set is an expensive proposition but RAM capacity tends to get more affordable over time, and RAM wasn't where we spent most of our money to begin with. The storage we selected will have enough capacity to last us for a while, and is more than fast enough for loading our operating system on initial boot-ups. If game files do continue to get larger as we projected, then it will begin to feel slow during loading screens and we will need to add more capacity. Both of those can be rectified by adding additional storage for our larger games in the form of M.2 PCIe 5.0x4 to our system in the future when storage with that speed becomes more common. The state of the graphics card market right now is not buyer friendly at all. We will save and wait until something we really want becomes affordable. Both the power supply and case should be fine to carry us through the decade or more we plan to use this system for.
Comments