Gaming: 8GB vs 16GB of RAM Tested

[Spotty]

(Tldr; Read the Conclusion and check Results for images & graphs)

 

Is 8GB of RAM sufficient for modern gaming?

 

 

Methodology
Run a mixture of in-game benchmarks and real gameplay.

Using FRAPS capture the Frametime data and plot the data on a graph using FRAFS comparing 8GB results to 16GB results. To use FRAPS to capture game Frametime Data you must go to the FPS tab within FRAPS and enable a hotkey for the benchmark tool. Set a timer for how long you want the benchmark tool to run. When the in-game benchmark begins, press the hotkey button to begin FRAPS data logging. Collected DATA is then exported and graphed using FRAFS.

Frametime data chosen as the primary metric of comparison as Average/Min/Max FPS data can fail to accurately reflect things such as microstutters that can be a result of memory issues.

Complete tests with minimal processes running in the background (closing all unnecessary background programs). The aim is to see the "Best case" scenario of what can be achieved with 8GB vs 16GB.

Completing additional tests with heavy memory load from background tasks - Achieved by opening chrome tabs in the background with up to 3GB of RAM usage for Chrome (clearly marked in results). The intention of these additional tests are to simulate results if people have programs opened in the background while they are gaming.

Pagefile is assigned to use the SSD (Samsung 840 Evo) when required.

Test System

CPU: Intel i7 6700k  (Stock 4.0GHz, TurboBoost disabled)

Motherboard: Gigabyte Z170x Gaming 5

RAM: 16GB (2x8GB) 3000MHz G.Skill Ripjaws 5 |    OR    |  RAM: 8GB (1x8GB) 3000MHz G.Skill Ripjaws 5

GPU: Gigabyte Aorus GTX 1080ti

PSU: Corsair HX750W

Case: BeQuiet SilentBase 800

Cooler: Corsair H100i AIO

SSD: Samsung 840 500GB | HDD: Seagate Ironwolf 8TB + 2x Seagate Ironwolf 6TB

Monitor: Acer Predator XB271HU

What is being tested

RAM Capacity - The difference between 8GB of RAM (1x8GB) and 16GB of RAM (2x8GB)

 

What is NOT being tested

RAM Speed (frequency, latency, timings, etc)
Dual Channel memory (ie. 2x4GB vs 1x8GB)

Different game settings (all the same settings were used for all tests)

Anything else

 

Games Tested

• Deus Ex Mankind Divided (DX11 and DX12 modes) - 2560x1440 Very High Preset, 2x MSAA
• Grand Theft Auto V "GTA V" - 2560x1440 Very High* settings (*See attached benchmark.txt files in additional notes for full graphical settings)
• Tomb Raider (2013) - 2560x1440 Ultimate preset
• Just Cause 3 - 2560x1440 Very High preset, Antistropicx16, SMAA_T2X, Motion blur off
• Metro Last Light Redux - 2560x1440 High Quality, SSAA On, Texture Filtering AF 4X, Motion Blur OFF, Tesselation High, Vsync Off, Advanced Phsyx On, Scene 1
• Middle Earth: Shadow of Mordor - 2560x1440 Very High preset, tesselation on
• PlayerUnknown's Battlegrounds "PUBG" - High preset {View Distance: Ultra}

 

The games were chosen due to their demanding nature on systems, mixture of game engines, popularity, availability of built in benchmarks, and by special request. Where possible, in-game benchmarks were used to ensure consistency between runs. Some actual gameplay results were also used.

Results

 

Deus Ex Mankind Divided (DX11)

In Deus Ex Mankind Divided (DX11 mode) there are only minor differences between 8GB and 16GB of RAM, with 16GB of RAM performing marginally better and achieving a slightly better FPS. Consistency is roughly equal between the two tests, and there are no noteworthy FPS drops or microstutters. Gameplay experience should not be severely affected, though there were minor improvements (few average FPS) seen when changing from single channel 8GB to dual channel 16GB RAM configuration. This may be a result of the change to using dual channel memory providing performance boosts, and it's possible that similar benefits could be seen when running 2x4GB (8GB) in dual channel memory configuration, although this scenario was not tested.

(Note: Ignore the spike and subsequent drop in the 8GB (orange) results at the end of the graph - This is a result of capturing data as the test concluded and should be discarded.)

 

Deus Ex Mankind Divided (DX12)
In Deus Ex Mankind Divided (DX11 mode) we see an interesting result where the results for each test slowly go out of sync of each other. The pattern seen in the graphs is a result of what images are being rendered on screen during the benchmark, and for one to go out of sync with the other indicates that one test is able to get to each section in the test earlier - ie, it is running the benchmark faster. This also could be due to an error in logging of data, however test was repeated and similar pattern formed again. This is something we will see repeated in another test later... I have absolutely no idea why it did this in DX12 mode test but not in the DX11 test. If anyone has any thoughts, please share them.
(I may end up repeating the tests again another day to rule out any errors in testing/graphing)

 

No noteworthy FPS drops or stutters to speak of, and results are fairly consistent again between 8GB and 16GB of RAM.
Gameplay experience on either systems will be comparable and perfectly enjoyable.

(Note: Disregard the spike and subsequent drop at the end of this graph for the 16GB test. This is a result of capturing data after the test was concluded and loading the results screen.)

 

Metro Last Light Redux
Metro Last Light was one of the tests that was extremely heavy on CPU and GPU, but in comparison relatively light on memory. Throughout benchmarking the benchmark didn't consume more than 2GB of system memory, meaning that at no time did the system need to rely on using pagefile. As a result, the two results between each test are near identical. There are quite a few small spikes throughout both tests, but appearing equally in both tests and often in the same areas. This could be an issue with the benchmark itself, or an issue caused on other hardware (CPU/GPU).

Gameplay experience with either 8GB or 16GB of RAM will be enjoyable

 

 

Middle Earth: Shadow of Mordor
This is one of our best results so far. We see a very consistent line between both tests with little variance. It also mostly stays below 8MS (above 125FPS+) which is a good result for the high resolution and graphical settings used. There are some small spikes on the 16GB of RAM test during the middle of the benchmark that are unexplained. These spikes may cause a noticeable stutter, however they are infrequent, possibly anomalies, so it's likely they won't be noticeable by the player or affect the gameplay experience. An extremely good gameplay experience can be had with either 8GB or 16GB of RAM.

(Note: The small higher grouping at the start, and the higher grouping at the end of the graph are recorded outside of the benchmark where a menu FPS limit is set - Disregard these sections)

 

Tomb Raider

Another fantastic result! Even better than the last. This is what a good Frametime graph should look like. Very consistent, little variance, absolutely no spikes, and a really low frametime (high FPS).
With either 8GB or 16GB you can expect an extremely smooth and enjoyable gameplay experience.

 

 

 

PlayerUnknown's Battlegrounds (PUBG)
PUBG is infamous for being poorly optimised. Compared to the Tomb Raider result above, it's obvious how it earned that reputation. This graph is a good example of a poor experience captured in Frametime data. There are many, many spikes in frametimes - resulting in microstutters noticeable to the player. There are also large chunks where framerate (FPS) drops. This is likely due to events happening within the game, such as explosions or being closer to other players, however it could also be caused by other factors including memory if it's struggling to load data properly.

This is captured gameplay ("Training Ground") data, and will naturally not be as consistent as the previous tests which used in-game benchmarks that can give repeatable results. As a result, it's hard to make any conclusions from the data. The 8GB test does seem to have a larger spread with less consistency, however the whole graph is such a mess that it's hard to make any solid conclusions between the two.

It is worth noting that system memory usage was high during this test, nearing the estimated ~7.5GB limit for 8GB of RAM before the system starts relying on pagefile. Due to this, it's extremely likely that issues will be worsened if you have background apps like browsers or music players open. My recommendation would be to aim for 16GB of RAM, however you're likely to have a choppy experience regardless of how much RAM you have.

 

 

 

Just Cause 3
Here we start to see more definitive impact that system memory has while gaming. This data is recorded during actual gameplay, as Just Cause 3 lacks a built in benchmark tool. To Just Cause's credit, if you launch the game with only 8GB RAM the system will provide a notice stating that you may experience some issues with lower system memory, and recommends closing background programs to improve gaming performance.

You can clearly see the 16GB test was able to maintain a lower frametime (higher FPS). There's also noticeable areas in the middle of 8GB test where there were periods of FPS drops for a few seconds - it's possible this is a result of reading from the pagefile and slowing as a result.

 

Since there's noticeable improvements to FPS and to smoothness, I would recommend 16GB system for playing Just Cause 3, though you can still have an okay experience with 8GB as long as you don't mind lower average FPS, the occasional stutters or FPS drops.

 

 

There were even some periods of high spikes, indicating very noticeable stutters in the game that were noticeable during gaming. To get a better look at how high some of those stutters were, we can increase the scale for the graph to from max 30ms to max 100ms. There are several noticeable spikes around 50ms, and even a spike higher than 100ms towards the start of the test - something that would be extremely noticeable to the player.

 

Since the game recommends you to close background programs for a better experience, I decided to test if having additional memory in use by background applications would worsen the situation.

It wasn't as bad as I expected it to be, but there are more noticeable chunks of FPS drops that were present in the previous 8GB test.

 

 

Grand Theft Auto V

This is the test I was most interested in seeing. "Stutters in GTA V" is one of the more common threads I see here, often caused by insufficient memory. So I was curious to see for myself how it performs in a system with only 8GB of RAM.

I was actually surprised by these benchmark results - in that the 8GB tests were a lot better than I was expecting them to be.

It was very interesting to see that the graphs went out of sync during this test, like we saw in the Deus Ex Mankind Divided DX12 test. In the case of GTA V, I believe it was a direct cause of the way the GTA V benchmark runs. There are 5 scenes which are benchmarked, with loading screens between each test. The system with 16GB of memory spent less time in the loading screen between tests, so was able to start each test earlier. As the tests went on and went through more and more loading screens, the difference between the two systems becomes more pronounced. From this result we can expect that having 16GB over 8GB will give a noticeable improvement in loading times in GTA V (which are pretty horrible, so any benefit that can be had we should take! Note: The game is installed on an SSD)

It's a little bit harder to see due to the graphs being out of sync, however the 16GB test did perform noticeably better than the 8GB test. In some scenes there was a difference of up to 10FPS between the two tests (see Average FPS result graph further below)

(Note: The flat bars across the bottom are loading screens inbetween each benchmark test scene. Loading times are shorter on the 16GB system.

 

The results from the first benchmark didn't show as significant performance issues with 8GB as what I had expected. I decided to load up a bunch of Chrome tabs again and put more stress on the system memory and run the test again, to see if that could impact performance further...

There was a further drop, with an average of up to 20FPS difference between the 16GB system and the 8GB system with background processes hogging memory.  The loading screens were significantly longer with this test as well with the graphs falling even further out of sync.

From this we can conclude the less background processes you have running, the better your experience should be.

 

 

Due to the nature of the GTA V benchmark with it running a series of short benchmarks, and then having a loading screen before the next test started, I was curious to see if actual gameplay would put additional load on the system as you remain in its open world for a longer period of time. I fired up Single Player mode and played for 5 minutes performing similar actions in each run, following a set path (Michaels house > Beach) and back and performing a variety of typical player actions such as driving, running, shooting, crashing my car many times, etc.

 

Here we see again the 16GB system performing significantly better than the 8GB system. We see approximately a 10-15FPS benefit from the 16GB system in actual gameplay. It should be noted that Task Manager was showing 97% memory usage and the system was definitely relying on page file during the 8GB test.

 

Again I tested with some background processes running, to simulate more of a 'worst case scenario' of someone having social media and a twitch video or youtube video playing music playing in the background while they play requiring additional memory from the system...

We see further performance loss here, similar to our earlier benchmark results.

 

I have also included the FPS results from the benchmark test. Included is the Minimum, Maximum, and Average FPS from each scene in the test, as well as the total average for the full benchmark test. The 16GB system performs consistently better than the 8GB system and the 8GB Stressed (with background programs running) system. In this graph the Average FPS is the best indicator of performance, as Max FPS and Min FPS can be effected by outliers. We can see a 10-15FPS drop

 

 

Conclusion
The results surprised me significantly. I went in to this fully expecting that most of these games would struggle significantly with only 8GB of RAM, however this was not the case.

Many of the games ran surprisingly well with only 8GB of RAM, with Tomb Raider, Middle Earth: Shadow of the Tomb Raider, Metro Last Light seeing no noticeable performance loss whatsoever. Deus Ex Mankind Divided saw minor performance loss in both DX11 and DX12 modes, however as long as background tasks are kept to a minimum there should be no noticeable loss in performance or smoothness in game.
PUBG shown what an absolute dog it is, and was equally bad in both 8GB and 16GB tests - Though my recommendation would still be 16GB for this game as I believe I was under a "best case scenario" where it was just able to copy with the RAM without needing to rely on the pagefile for system memory - Your experience may vary.

Just Cause 3 and GTA V showed more significant performance drops as a result of the memory, especially when background programs were running such as open browser tabs.

If you're building a gaming PC on a budget, then for the most part 8GB will be okay to start with, though you should still plan to upgrade to 16GB further down the line as there are games that will see benefit from the additional RAM.

If you're playing games like GTA V and Just Cause 3 with a system with only 8GB of RAM and are unable to upgrade to 16GB, my recommendation is close as many background processes as you can while you are playing, and try to disable as much useless programs from running at startup as possible in order to free up as much system memory as possible.

Additional Information

 

GTA V Benchmark Results: (Contains full graphical settings data)
8GB Benchmark-18-10-26-14-17-34.txt
8GB Stressed Benchmark-18-10-26-16-45-14.txt

16GB Benchmark-18-10-26-13-44-19.txt

 

Unless otherwise stated, the graphs used in the results section have their scale set from 0ms to a maximum of 30ms. This demonstrates the results with the best accuracy and ensures all results can be easily compared. Any frametimes measured beyond the 30ms point will simply hit the top of the chart. In some cases charts with a scale of 0ms to a maximum of 100ms were used to better demonstrate larger frametime spikes (stutters) - these graphs are noted as "(100ms max)".

 

 

Why did you choose 1x8GB instead of 2x4GB? You won't get dual channel!

I chose to use 1x8GB instead of 2x4GB for two main reasons.

1. I often see recommendations in budget build threads of people recommending to start out with 8GB of RAM, typically a 1x8GB stick, to keep the initial build cost down and then adding another 8GB later on. I wanted to test the viability of this plan.
2. I don't have any 4GB DDR4 sticks

From what I've seen in some of the games that weren't running out of memory, such as Tomb Raider and Middle Earth: Shadow of Mordor, the final results between 8GB (1x8GB) and 16GB (2x8GB) ended up being extremely close, almost indistinguishable. So it appears any performance loss from using single channel memory will be minimal in gaming tasks.

 

 

How do I interpret all those graphs? What do they all mean?

"LOWER IS BETTER, BUT CONSISTENCY IS EVEN BETTER STILL" - Steve, Gamers Nexus

Frametime is measured in milliseconds and is the time in between each frame. For 60FPS, there is 16.67ms between each frame. For 144FPS, there is 6.95ms between each frame.

When it comes to Frametime; LOWER IS BETTER. However, CONSISTENCY is also extremely important. Ideally the graph should display a low Frametime as well as a consistent line with as little deviation as possible. If a graph shows large spikes in frametime, this can be an indication of stutter. Some small spikes may not be noticeable, however larger spikes can cause a noticeable stutter in game which appears as if the game 'freezes' for a fraction of a second.

 

 

Below is an example of a GOOD frametime graph. Note the tight grouping of data and consistent nature. Very little variance, and the vast majority of the frametime is measured at below 8ms (125FPS). There are no frametime spikes indicating stutters or groupings of frametime spikes indicating framerate drops (low FPS).

Spoiler

 

Below is an example of a BAD frametime graph - Note the inconsistent nature with how the graph is more spread out indicating less consistency in FPS, as well as areas where there are large groupings of spikes marked red - resulting in noticeable framerate drops (lower FPS). Also note the single points marked green where the frametime peaks quite high which can be an indicator of a short stutter in game.
Note, the area marked in red with groupings of higher frametimes are often the result of whatever is happening in the game at the time, such as, for example, when large explosions take place that result in a small FPS drop while the game is displaying the explosion on screen and performing the physics calculations.

Spoiler

 

 

So, bad Frametime is caused by insufficient memory?
NO! Frametime is determined by your entire system, including the CPU and Graphics card. If we were comparing a weaker graphics card against a stronger graphics card, we would expect the weaker graphics card to have consistently higher frametimes than the stronger card.

It's important to use Frametimes to detect issues such as microstutters, as single frames that take longer to render and display may not be accurately reflected in average FPS calculations. Since the only variable we are changing is the RAM capacity used, with all the other hardware such as CPU and GPU remaining the same, we can deduce that any changes between the two results (8GB vs 16GB) are a result of the change in memory.

 

Limitations & Errors in Testing

• In-game benchmarks used for most tests - May not accurately reflect real gameplay
• When actual gameplay was used, there is natural variance in RAM usage and system demand based on what tasks are happening within the game.
• Memory in 8GB configuration was not running in Dual Channel mode, which may give a slight advantage to the 16GB (2x8GB) configuration which was running in Dual Channel mode.
• Limited games tested. - I can only test what games I own.
• "My favourite game XYZ wasn't tested, it uses heaps of memory!!!!" - I can only test what games I own. If you would like me to test a specific game then please feel free to request it, however I cannot guarantee I will be able to test it.
• This is the first time that I've focused on Frametime (ms) as the primary metric for benchmarking, and there may be some inconsistencies in reporting
• RAM Usage & Pagefile Usage wasn't logged - I would have loved to be able to log the RAM/Pagefile usage alongside the Frametime data, however I'm not aware of a way to accurately log such information in a way that can be imported in to the Frametime Graph.
• Only 2560x1440 resolution, and higher graphical settings were used. Different resolutions, such as 1080p or 4k, as well as different graphical settings may impact results.
• "The bottom axis is sometimes wrong or has the numbers doubled up! FAKE NEWS!" - This is a result of the way FRAFS graphs its data. The axis is labelled with the FRAMES processed instead of DURATION (seconds) of the test.
  Often benchmarks are set to run for a certain run, and if one test is running faster than the other then the faster test will complete more frames. (For example, 1 test running at 60fps for a minute will run 3600 frames. Another test running at a much faster 120FPS will complete 7200 frames in the same time. This means that the scale labelling frames at the bottom might not be synced between each test, however the time each test ran for was consistent)
  The two result graphs were overlayed over the top of each other to obtain the final comparison graphs you see here, as a result some of the frames labelled at the bottom of the scales do not align. Unfortunately there's no way to label the axis with the TIME (duration) of the test, which would remain consistent between each run. Ultimately my advice is to ignore the labels on that bottom axis. (I couldn't be bothered removing them).
• Extra stress on memory was achieved by opening tabs in Chrome during some tests (clearly noted which). This may have put additional stress on the CPU as well, which may have affected the results. These results are only provided in addition to the other results and do not replace them.

If you notice any other errors, please comment below and I'll review them and add them to this list.

 

 

[Spotty]

[Reserved]

 

Spoiler

May update with additional tests if required

 

[Mira Yurizaki]

Nice job!

 

One comment: even if Chrome or other apps are eating up all that memory, it'll have to be ejected into the page file. So you'll only have lag spikes during that swap. Unless of course Chrome is like "oh no no no you don't, I want that RAM back" and tries to do something to get the memory manager to put it back in.

 

The only other thing I would've liked to see was RAM usage over time.

[genexis_x]

Hmm...8GB surprises me despite running single channel

[Spotty]

1 minute ago, M.Yurizaki said:

One comment: even if Chrome or other apps are eating up all that memory, it'll have to be ejected into the page file. So you'll only have lag spikes during that swap. Unless of course Chrome is like "oh no no no you don't, I want that RAM back" and tries to do something to get the memory manager to put it back in.

I did notice that...
To get the worst case scenario, I opened a bunch of tabs and was able to get Chrome memory usage up to 3GB, though once I fired up the game it seemed it offloaded a lot of Chrome since it was a low priority background processed and it went down to ~1GB - 1.5GB.
So it seems Windows handles itself pretty well in that regard, though there was still a noticeable drop in performance between the tests when I did that.

 

3 minutes ago, M.Yurizaki said:

The only other thing I would've liked to see was RAM usage over time.

I would have loved to add it as well. Especially just a flag at the graph at any time it ran out of available RAM... But it would have meant combining data from multiple different monitoring software, which can have their own variances, and would have also been a nightmare to integrate in to the graphs properly. This was the first time measuring primarily with frametime so maybe in the future I can work out a better system that can incorporate that data...
... I was keeping a very close eye on it when I was running the benchmarks. Tomb Raider, Middle of Earth, and Metro didn't come anywhere close to reaching 8GB total system memory. GTA V & Just Cause 3 were either maxing out or close to maxing out that available RAM with 8GB installed. Obviously no concerns with any of the games with 16GB.

[Spotty]

4 minutes ago, GeneXiS_X said:

Hmm...8GB surprises me despite running single channel

I suspect the minor differences between Deus Ex results *may* have been a result of single channel vs dual channel mode.

I initially intended these tests to be something I could link to in order to convince people to spend the extra cash to buy 16GB of RAM instead of 8GB. I was certain the 8GB tests would tank hard... But I guess I was wrong...

I forgot to mention it in my post, but I chose 1x8GB instead of 2x4GB for two reasons...

1. I see people often recommending getting 1x8GB stick and then add another 8GB stick later when recommending budget builds, so I wanted to test it to see if it is a viable option and figured that due to the recommendations it might be a realistic approach someone will take...
2. I don't have any 4GB DDR4 sticks

[glenalz81]

I built my gaming rig when RAM was pretty much at it's peak price, chose a dual channel 8gb  kit with the intention of adding to it later but to be honest my max RAM usage playing any games i play is about 6.2gb in one title, other than that it's 4.5/5.5gb when gaming and i've never felt the need to add more.

[Mira Yurizaki]

1 minute ago, Spotty said:

I would have loved to add it as well. Especially just a flag at the graph at any time it ran out of available RAM... But it would have meant combining data from multiple different monitoring software, which can have their own variances, and would have also been a nightmare to integrate in to the graphs properly. This was the first time measuring primarily with frametime so maybe in the future I can work out a better system that can incorporate that data...
... I was keeping a very close eye on it when I was running the benchmarks. Tomb Raider, Middle of Earth, and Metro didn't come anywhere close to reaching 8GB total system memory. GTA V & Just Cause 3 were either maxing out or close to maxing out that available RAM with 8GB installed. Obviously no concerns with any of the games with 16GB.

Windows has a built-in tool for for logging a lot of system parameters. I think it was called perfmon or something. The only problem would be trying to line up the graphs unless you consistently do something like "press start, count to 30, then start the benchmark"

 

3 minutes ago, GeneXiS_X said:

Hmm...8GB surprises me despite running single channel

 I've seen this before (https://www.hardwaresecrets.com/does-dual-channel-memory-make-difference-in-gaming-performance/) so seeing this being observed again is nice.

 

While it may be reasonable to assume that higher bandwidth available should equal higher performance, I have no reason to believe that applications actually need that bandwidth most of the time.

[sjb655]

Thanks for doing this, was really helpful

[NoxiousOdor]

On 10/26/2018 at 9:38 AM, M.Yurizaki said:

While it may be reasonable to assume that higher bandwidth available should equal higher performance, I have no reason to believe that applications actually need that bandwidth most of the time.

Not only this, but most games in my experience don't even mind slow ram speed. After a goofy overclocking mistake I once found that I had been running my ram at 600mhz for quite a while without a significant difference in game performance. 

[Spotty]

1 minute ago, WhisperingKnickers said:

Not only this, but most games in my experience don't even mind slow ram speed. After a goofy overclocking mistake I once found that I had been running my ram at 600mhz for quite a while without a significant difference in game performance. 

It will depend on the CPU. Ryzen CPUs are more sensitive to RAM speed.

 

Quote

The Infinity Fabric: A Blessing And A Curse
There is much speculation (and plenty of proof) that faster memory improves Ryzen's gaming performance. The theory is that the speed of AMD's Infinity Fabric is tied to the memory clock rate, and through a bit of our targeted testing below, it looks like this is true.

 

https://www.tomshardware.com/reviews/amd-ryzen-5-1600x-cpu-review,5014-2.html

 

[Cereal5]

(This is the first review I read, I didn't even know we had this section lol)

Not to rain on a parade, but I'm gonna have to disagree with the methodology here. I almost always recommend a 2x4GB kit over a 1x8 kit due to the limitations of single channel RAM as opposed to dual or quad channel. While you may not have run out of memory in some tests, you very well could have run out of bus width, which would liken to the CPU sending a large number of data requests, but not actually large data. This would lead to a slow down in the system even though it says you are not out of memory. You can see this mainly in the GTAV benchmark. Also, your testing in Just Cause 3 fails to mention any memory usage nearing 8GB, so I can only assume you had these problems while not using your maximum RAM allotment.

 

Assuming all other constants, and the system not using the full 8GB of RAM, this is the only explanation to slower speeds.

 

You can also see this in Deus Ex DX11. You mentioned that Deus Ex, Tomb Raider, and Middle Earth. Also your difference in graphs in Deus Ex DX12, again assuming no other factors, would point only to a single channel bottleneck. In sections on Tomb Raider you see slightly higher frametimes around the first 1/3 and second 1/3.

 

I do see you mentioned this towards the end, as well as in the Limitations section. I will say I'm surprised at how close the two configurations are. I always thought single channel was slightly worse than even this. It would be interesting to see the same tested in CPU tasks. Thanks for the research though! Very well written piece

[Spotty]

4 minutes ago, Cereal5 said:

Not to rain on a parade, but I'm gonna have to disagree with the methodology here. I almost always recommend a 2x4GB kit over a 1x8 kit due to the limitations of single channel RAM as opposed to dual or quad channel. While you may not have run out of memory in some tests, you very well could have run out of bus width, which would liken to the CPU sending a large number of data requests, but not actually large data. This would lead to a slow down in the system even though it says you are not out of memory.

This is true. When tested in artificial tests and CPU tests, the memory in single channel vs dual channel memory played a larger role. I think the Cinebench scores (not included in above review) were about 80-90pts lower on the single channel 8GB stick than with the dual channel (2x8GB) 16GB configuration.

 

16 minutes ago, Cereal5 said:

Assuming all other constants, and the system not using the full 8GB of RAM, this is the only explanation to slower speeds.

  

You can also see this in Deus Ex DX11.

Something I didn't discuss in the thread, but something I've found in previous testing, is that Deus Ex Mankind Divided in particular is sensitive to memory clock speeds, timings, and as well as obvious things such as CPU clocks. I think the difference between 2666Mhz and 3000Mhz in Deus Ex resulted in an average 3-5FPS difference, but it's been a long time since I've tested that and I don't have the results on hand to confirm...

 

It's entirely likely that the minimal performance difference seen in between the two Deus Ex tests is due to single channel vs dual channel. However, it's worth mentioning that this small performance drop as a result of this would not significantly adversely effect the gaming experience one would receive, and outside of benchmarks the player would likely not notice a consistent 1-3fps average drop once all other things such as different CPUs, GPUs, memory frequencies and timings, etc were all accounted for. One of the primary things I was looking for in this test were significant performance drop and stuttering caused directly as a result of insufficient memory capacity.

 

Something that I did note was that the 16GB configuration gave slightly better performance, although now re-reading it I do realise the way I have worded it is misleading in saying that the change in capacity to 16GB was the result for the slightly higher FPS, rather than the change to dual channel 2x8GB being the result, so I will update that to better reflect the results.

 

On 10/27/2018 at 3:13 AM, Spotty said:

Consistency are roughly equal, and there are no noteworthy FPS drops or microstutters. Gameplay experience should not be severely affected, though there may be some minor improvements (few avg FPS) if upgraded to 16GB of RAM.

Will update to "Consistency are roughly equal, and there are no noteworthy FPS drops or microstutters. Gameplay experience should not be severely affected, though there were minor improvements (few average FPS) seen when changing from single channel 8GB to dual channel 16GB RAM configuration. This may be a result of the change to using dual channel memory providing performance boosts, and it's possible that similar benefits could be seen when running 2x4GB (8GB) in dual channel memory configuration, although this scenario was not tested" to better reflect the results and testing.

 

17 minutes ago, Cereal5 said:

While you may not have run out of memory in some tests, you very well could have run out of bus width, which would liken to the CPU sending a large number of data requests, but not actually large data. This would lead to a slow down in the system even though it says you are not out of memory. You can see this mainly in the GTAV benchmark. Also, your testing in Just Cause 3 fails to mention any memory usage nearing 8GB, so I can only assume you had these problems while not using your maximum RAM allotment. 

GTA V and Just Cause 3 were both definitely out of memory during the tests performed. From memory with Just Cause 3 running the system memory usage sat at around 7.5-7.7GB memory usage during the 8GB tests with the pagefile being utilised. With the 16GB of tests the system sat at around 8.5-9GB of memory usage. The open world games definitely relied more heavily on memory usage than the more linear, closed world games, which does make a fair bit of sense - however my testing was too limited in its choice of games for me to confidently conclude that open world games would require more RAM.

 

27 minutes ago, Cereal5 said:

It would be interesting to see the same tested in CPU tasks.

I did some extremely basic and limited CPU testing, such as Cinebench, but I wanted to focus primarily on Gaming tests, and my post was already quite long and information heavy so I didn't want to overwhelm anyone with too much secondary information. Since you're curious... With the CPU in the state it was used for the testing (locked to 4GHz, Turboboost disabled), Cinebench tested with 8GB scored around 720 - 730pts, and about 810-820 pts for 16GB (2x8GB dual channel).

 

Cinebench testing (8GB is highlighted result (726), 16GB tests are the two directly above it (810, 818). Ignore the higher cinebench scores as they are unrelated tests from CPU overclocking tests and have no relation to this testing whatsoever)

Spoiler

(Don't take these results too seriously, as I was just kind of firing it up and running it while I was doing other things just to see what the results would be for my own curiosity. Having all those background processes running would probably cause variance in the results. I was not intending to use these results in my review so I was not controlling the variables as well as what I would have under normal testing situations)

 

32 minutes ago, Cereal5 said:

I do see you mentioned this towards the end, as well as in the Limitations section. I will say I'm surprised at how close the two configurations are.

My aim is to be as transparent and open in my testing as possible, and identify any limitations or errors in order to help ensure that I do not misrepresent the results...

 

Also:

Quote

Why did you choose 1x8GB instead of 2x4GB? You won't get dual channel!

I chose to use 1x8GB instead of 2x4GB for two main reasons.

1. I often see recommendations in budget build threads of people recommending to start out with 8GB of RAM, typically a 1x8GB stick, to keep the initial build cost down and then adding another 8GB later on. I wanted to test the viability of this plan.
2. I don't have any 4GB DDR4 sticks

 

If I had 2x4GB DDR4 sticks on hand, I'd probably test that configuration against 1x8GB in addition to this testing... But I'm not going to spend AUD$150 on RAM just for the sake of a forum post

The whole debate of 2x4GB vs 1x8GB is probably its own debate to be having, and if I were to do it I would dedicate a thread for it specifically and look at a wider range of tests, such as CPU benchmarks, gaming, rendering and editing, etc.

[Cereal5]

6 minutes ago, Spotty said:

It's entirely likely that the minimal performance difference seen in between the two Deus Ex tests is due to single channel vs dual channel. However, it's worth mentioning that this small performance drop as a result of this would not significantly adversely effect the gaming experience one would receive, and outside of benchmarks the player would likely not notice a consistent 1-3fps average drop once all other things such as different CPUs, GPUs, memory frequencies and timings, etc were all accounted for. One of the primary things I was looking for in this test were significant performance drop and stuttering caused directly as a result of insufficient memory capacity.

Yeah, I'd agree with that. That's part of the reason why I said I was surprised at how small the difference was.

 

7 minutes ago, Spotty said:

Consistency is roughly equal (or "consistencies are")

FIFY  sorry I'm that kinda person lol

 

9 minutes ago, Spotty said:

The open world games definitely relied more heavily on memory usage than the more linear, closed world games, which does make a fair bit of sense - however my testing was too limited in its choice of games for me to confidently conclude that open world games would require more RAM.

I would definitely agree with you there. My experience in open-world type games is very limited though, and none have I played recently.

 

10 minutes ago, Spotty said:

I did some extremely basic and limited CPU testing, such as Cinebench, but I wanted to focus primarily on Gaming tests, and my post was already quite long and information heavy so I didn't want to overwhelm anyone with too much secondary information. Since you're curious... With the CPU in the state it was used for the testing (locked to 4GHz, Turboboost disabled), Cinebench tested with 8GB scored around 720 - 730pts, and about 810-820 pts for 16GB (2x8GB dual channel).

Taking into consideration your lack of control for the runs, I'd still say that's pretty significant. Even shrunk to 50 points, that's 12%+ off the 16GB score, and I know cinebench doesn't use 8GB of RAM. I wonder how triple and quad channel fares....... If only my X99 board didn't DIE ON ME!

 

13 minutes ago, Spotty said:

If I had 2x4GB DDR4 sticks on hand, I'd probably test that configuration against 1x8GB in addition to this testing... But I'm not going to spend AUD$150 on RAM just for the sake of a forum post

I don't blame you, nor would I lol. I might if they threw something towards those that did... Then I have a reason to hoard PC parts LOL.

[Spotty]

2 minutes ago, Cereal5 said:

FIFY  sorry I'm that kinda person lol

This is what happens when you rewrite a sentence 5 times.

 

4 minutes ago, Cereal5 said:

Yeah, I'd agree with that. That's part of the reason why I said I was surprised at how small the difference was.

I was pretty surprised how well 8GB held up in most of the games. I was expecting all of the games I tested to fail miserably, and then I was planning on following it up with tests of a bunch of lighter esports titles such as CSGO, LoL, Overwatch, Fortnite, etc to see if the lighter games would be playable on 8GB of RAM. I proved myself wrong

 

7 minutes ago, Cereal5 said:

Taking into consideration your lack of control for the runs, I'd still say that's pretty significant.

It's definitely outside the margin of what would be caused by tasks like Firefox running in the background, so I'd say single channel configuration would be respsonible for a considerable portion of the performance loss. I can't remember as it was a few weeks ago, but I probably had a bunch of programs open like firefox when I did the 16GB testing anyway - As I mentioned, I wasn't intended to use those results in my review I was just curious what would happen. I just ran cinebench again with 16GB, while having Firefox, discord, steam, etc open and I got a cb score of 813pts. CPU is still at stock 4.0GHz cause I haven't bothered changing it back since I did my testing, plus it's 37°C here today and we're starting to head in to Australian summer, so overclocking the CPU isn't really much of an option for the next 6 months or so...

Cinebench only uses about 600MB of memory, so the capacity isn't an issue but it's how quickly the CPU can access information from the memory that is important in Cinebench. Looks like single channel memory can account for more than 10% performance loss in heavily CPU & memory bound tasks, which is fairly significant... Though, that's probably a discussion for another thread...

[Cereal5]

5 minutes ago, Spotty said:

I was pretty surprised how well 8GB held up in most of the games. I was expecting all of the games I tested to fail miserably, and then I was planning on following it up with tests of a bunch of lighter esports titles such as CSGO, LoL, Overwatch, Fortnite, etc to see if the lighter games would be playable on 8GB of RAM. I proved myself wrong

Yeah the difference in eSports titles probably doesn't even matter, cause your monitor won't be able to push out that many frames anyways lol. (OVER 9000 in the case of CSGO).

 

6 minutes ago, Spotty said:

It's definitely outside the margin of what would be caused by tasks like Firefox running in the background, so I'd say single channel configuration would be respsonible for a considerable portion of the performance loss. I can't remember as it was a few weeks ago, but I probably had a bunch of programs open like firefox when I did the 16GB testing anyway - As I mentioned, I wasn't intended to use those results in my review I was just curious what would happen. I just ran cinebench again with 16GB, while having Firefox, discord, steam, etc open and I got a cb score of 813pts. CPU is still at stock 4.0GHz cause I haven't bothered changing it back since I did my testing, plus it's 37°C here today and we're starting to head in to Australian summer, so overclocking the CPU isn't really much of an option for the next 6 months or so...

I have multiple runs of my CPU at stocks speeds. I don't remember how the first test was done, but I got a score of 1337. No joke. The second test (a few days later) I scored a 1351, with Cinebench, task manager, and CPU-Z up, and a few softwares (corsair link, afterburner, logitech, etc) running in the background. I tested again right after after closing all of the non essentials, and got a 1370. Previous consecutive runs indicate that the difference in score was most likely more than just average deviation. I don't know, it's just interesting to learn niche things like this, how even something that task manager shows as taking up 0% CPU usage can still create small effects like that.

 

10 minutes ago, Spotty said:

Though, that's probably a discussion for another thread...

Yeah, I think that's where I should end lol without discussing a whole new thread's worth of info in the wrong thread.