We Raise the Question Once Again
About this time each year we set out to see how much RAM is required to play the latest and greatest PC games. Last time we concluded that 8GB should be the minimum standard, while 16GB was desirable but not truly needed for general usage or gaming. Now a year later, we suspect this is still true, but we have seen a few games sneaking past 8GB of allocated memory so it's worth looking into again.
Something that stings a little is this comment from last year's article: "...system memory has hit new lows in 2016 which makes it very accessible even in budget builds." That's no longer the case, of course, so we suspect those of you with 8GB of memory will just make do until later next year when prices hopefully come down.
On the last go around we looked at a number of applications and found memory usage varies massively, so what you're doing with your PC will ultimately decide just how much memory you need. For basic office type work, 8GB should be more than enough. For content creation, the more the merrier -- 16 GB for example is the bare minimum when working with 4K footage in Premiere Pro.
It's hard to determine the impact that RAM capacity has on gaming because of all the factors at play. Before we get into the results, we'd just like to discuss a few challenges we faced when testing system memory capacity.
The graphics card you use can influence how much memory you'll need for smooth gameplay, the speed of your storage can also impact performance.
The first and possibly the biggest challenge is picking the right hardware to test RAM capacity. The graphics card you use can influence how much memory you'll need for smooth gameplay, the speed of your storage can also impact performance.
For example, a GTX 1060 3GB will often see higher system memory usage than a 6GB model because when the former runs out of VRAM it uses system memory. If you run out of system memory, some game assets are moved to the local storage device, which is hopefully an SSD, and depending on how fast that drive is and how heavily it's hit with data, you may or may not see a noticeable dip in framerate.
So, testing with a GTX 1080 Ti and its 11GB VRAM buffer along with a super snappy NVMe SSD probably isn't the best way to determine how much system memory your average gamer requires. We'd also assume that if you have a $700 graphics card and a primary storage device of a similar value, buying 16GB or more of RAM isn't really an issue.
There are other factors to consider as well. Memory speed can influence the results, granted even the fastest memory, think HBM, can't make up for a lack of capacity but in some instances when working with large volumes of broken up data it can help to avoid slow downs as the data is moved in and out of system memory much quicker.
The quality settings used in games can also really impact the results. If you use a GTX 1060 3GB but are happy turning things like textures down to a medium quality setting then that will reduce how much data is offloaded to the system memory.
Taking all this into account, we've tested a number of different hardware configurations and we'll be showing the impact all this has on performance. However, we quickly discovered another issue with trying to show the fps difference between the various hardware configurations. Benchmark passes typically last anywhere from 30 to 60 seconds and are run multiple times. This means the system has a chance to cache the pass and only keep the data we're constantly in high speed memory.
So, while the results from the first run might see a shockingly low 0.1% or 1% frame time figure, this can be improved dramatically on the second run and then again on the third. Displaying the three run average can be misleading.
The best solution we came up with for addressing this problem was to run each benchmark once, then reset the entire system, load the game again and run the next pass. This means the 30 second pass doesn't have a chance to be cached.
The benchmark pass is actually 60 seconds in total but we only report the frame rates for the last 30 seconds. This mimics more what you'd see when actually playing the game as the software can still be loading assets for the first 5-10 seconds that you are playing, and this can cause frame drops, even on computers with sufficient system memory.
Typically, the solution to this problem is to run the benchmark pass once, throw away those results and then run three more times, at which point you take the average from those three runs. That method isn't useful for measuring the impacts of memory capacity though.
Finally, benchmarks aside, a good indicator for working out how much memory you'll need to play the latest games is to just monitor memory allocation. It's not foolproof but it gives us a pretty good idea of how much system memory a game requires to avoid any slow downs. Again the hardware used will impact the amount of system memory used.
Having said all that, we're first going to show you memory usage using three different graphics cards in a few popular multiplayer titles, all these gameplay tests were conducted with a Core i7-8700K with 32GB of DDR4-3000 memory. The test system has a number of applications running in the background such as Steam, Origin, Uplay, Discord, Chrome with a few tabs open, MSI AfterBurner, RivaTuner and Fraps.
So it's by no means a completely clean test system running a skeleton crew of applications. Also please note that the maps used aren't always the same so this isn't an exact comparison, but the memory usage does reflect what we saw for the most part.
Starting with a 64-player Battlefield 1 match, we first tested on the GTX 1080 Ti at 1440p with the ultra quality preset and found that it used up to 8.2GB of system memory. The system page file grew to 11.6GB in size and this resolution/quality combo sees 3.8GB worth of VRAM consumed -- the 1080 Ti of course has an 11GB buffer. This suggests that for perfectly smooth performance in this title gamers will require 16GB of system memory.
Dropping down to the GTX 1060 6GB under the same test conditions, we see RAM usage peak a little higher at 8.5GB and the system page file grew ever so slightly to 12GB. Again, we're seeing around 3.6GB worth of VRAM allocation.
Now, what if we use a graphics card with a VRAM buffer that's smaller than the amount of memory the game wants to allocate? With the GTX 1060 3GB we see RAM usage hit 10GB while the pagefile grows to 12.5GB and we saw occasional dips in the frame rate as a result.
That said, gameplay was for the most part smooth thanks to the massive 32GB of system memory we installed. But whereas you might have gotten away with 8GB using the GTX 1080 Ti, you most certainly require 16GB with a graphics card sporting less than a 4GB framebuffer.
However, the 3GB 1060 is really intended for 1080p gameplay but even so lowering the resolution to 1080p still sees RAM usage peak at 9.7GB, though for the most part the game only allocates around 2.6GB of VRAM.
I'll quickly summarize what I saw in the rest of the games... Running PlayerUnknown's Battlegrounds on the GTX 1080 Ti saw 7.5 GB of system memory used and throughout my testing at 1440p@ultra quality, VRAM usage peaked at 6.2GB.
The 6GB GTX 1060's system memory allocation climbed to 8.3GB while the pagefile grew to 14.5GB.
With the 3GB GTX 1060 installed, RAM allocation boiled over to 12GB and reducing the resolution to 1080p didn't help as we still hit 12GB.
Call of Duty WWII
Now for the ultimate memory pig, Call of Duty WWII. Using the extra quality settings at 1440p we see the GTX 1080 Ti VRAM usage peaking at 10.8GB and that saw RAM allocation hit 10GB -- wow.
Interestingly, RAM usage with the 6GB GTX 1060 was much the same, as was the page file size. The comparison unfortunately wasn't made on the same map but I did try multiple maps and saw similar usage.
Even with the 3GB GTX 1060 we see similar system memory usage but frame drops became very apparent. Dropping down to 1080p didn't help and gamers using the 3GB 1060 will need to lower attributes such as texture quality from extra to something less demanding, maybe normal would be appropriate.
Overwatch isn't known for it's super high-quality textures, even when using the epic quality settings. Here we see at 1440p, the GTX 1080 Ti only saw up to 2.4GB of VRAM usage while system memory hit just 6.9GB, suggesting that 8GB would be fine for this hardware configuration in Overwatch.
We see similar usage patterns with the GTX 1060 6GB. System memory usage peaked at 7GB so again, 8GB of RAM will work well for this title.
With the 3GB 1060 you're right on the edge as RAM allocation hit 7.5GB in my test and much the same was also seen when testing at 1080p.
Star Wars Battlefront II
Finally, I monitored memory usage using Star Wars Battlefront II, and no I didn't find any extra system memory in a loot box. System memory usage hit 9.2GB when using the GTX 1080 Ti at 1440p with the maximum in game quality settings, so this is yet another title that will benefit from 16GB of RAM.
Oddly dropping down to the 6GB 1060 reduced memory allocation to a peak of 8.5GB, I did re-test this a few times and found the same result each time.
With the 3GB 1060 installed, RAM allocation shot up to 10.4GB, suggesting that with this graphics card you'll certainly want 16GB of system memory. Even at 1080p, usage still rose above 10GB.