Single Stick vs. Dual Channel RAM: How Much Performance Do You Actually Lose?

[dakta]

I'll join the club and thought I'd be seeing a comparison vs single and dual channel for a similar amount of memory. (1 x 16gb or 2 x 8 gb).

Doubling the ram makes for less consistency and I don't really know what to take from it, where is the difference being made?

I always kinda knew more RAM is good until you have too much when the benefits drop off a cliff, but I do have a system due to the incremental nature of my upgrades with dissimilar quantities and I wondered if there'd be a tangible difference and how much.

Maybe another time....

 

[Shirley Dulcey]

That's not how it works. You don't gain performance just by having a larger amount of RAM installed in your system. With all else equal, performance between 16 GB and 32 GB of RAM is identical unless you exceed 16 GB of RAM in active use, which shouldn't be the case in any of the games tested here.

It's baffling that so many of you here, on a tech enthusiast forum, don't understand this.

There is always SOME use of the additional RAM. If the game itself doesn't need it, Windows (or any other modern OS) will use the rest as a disk cache, and that can also improve performance. I agree that it would have been a more fair comparison if they kept the amount of RAM equal.

When single channel memory REALLY kills you is if you are using an iGPU. You're not going to be gaming on the iGPU in most desktop processors, except for possibly an AMD APU with a G at the end of its model number; those are really laptop processors in a desktop package. But it can be a huge consideration on laptops.

 

[poshflamingos]

There is always SOME use of the additional RAM. If the game itself doesn't need it, Windows (or any other modern OS) will use the rest as a disk cache, and that can also improve performance. I agree that it would have been a more fair comparison if they kept the amount of RAM equal.

First of all, no, there's isn't "always some use of additional RAM". If that were true, windows would immediately consume 100% of RAM at boot.

Second, windows does not use RAM as a "disk cache", that's not a thing. What windows does have is a feature called prefetch, which will preload executables/libraries for a few programs that you use frequently so they open faster. That is 1) not a disk cache; 2) limited to a small portion of your RAM, and freed up if necessary; and 3) something that doesn't affect CPU or GPU performance in any way whatsoever. The only thing it does is make those specific programs launch a little faster, it doesn't affect gaming performance/framerate at all.

Again, it's baffling how many of you make those vague, confidently incorrect claims about things you clearly do not understand well enough.

 

[Nima304]

I agree with the commentors saying that systems in this test likely weren't using >16GBs of RAM, so going to 32GB for the dual-channel test likely isn't an issue. However, doubling the RAM in such a test is bad science, and I'd like to see 1x32GB test results to remove all doubt.

 

[gamerk2]

I'm not thrilled with this test; if you really wanted to test single-vs-duel then you should have tested 2x16 versus 1x32, so there isn't any RAM bottleneck and you actually test bandwidth.

 

[TheMightyPeng]

First of all, no, there's isn't "always some use of additional RAM". If that were true, windows would immediately consume 100% of RAM at boot.

Second, windows does not use RAM as a "disk cache", that's not a thing. What windows does have is a feature called prefetch, which will preload executables/libraries for a few programs that you use frequently so they open faster. That is 1) not a disk cache; 2) limited to a small portion of your RAM, and freed up if necessary; and 3) something that doesn't affect CPU or GPU performance in any way whatsoever. The only thing it does is make those specific programs launch a little faster, it doesn't affect gaming performance/framerate at all.

Again, it's baffling how many of you make those vague, confidently incorrect claims about things you clearly do not understand well enough.

I created an account purely to respond to your haughty tone and the rubbish you have posted. Windows does in fact use additional RAM if it is present - up to a point.

"If that were true, windows would immediately consume 100% of RAM at boot."

I had to stop reading for a minute after that gem. Absolutely hilarious.

The preamble to this article clearly states "...do we really need two sticks? After all, most games still run fine with 16 GB of RAM. So how much performance are we sacrificing when going from dual-channel to single-channel memory? Today, we're going to answer exactly that."

Therefore the stated premise of the test is clearly to establish the performance difference, if any, between running in either single-channel or dual-channel and the ONLY valid way of doing that is to keep the amount of RAM the same e.g. 1 x 16GB stick (single-channel, obviously) vs 2 x 8GB sticks (dual-channel). Throwing in the extra variable of differing RAM amounts renders the clearly-stated purpose of the test invalid.

I have no idea what you do for a living but personally I've been working in IT Support (started as a technician, now a network manager) for 30 years in which time I have built/tested/diagnosed/upgraded thousands of systems, and as such I am supremely confident in my level of technical expertise. Anybody with even a modest amount of diagnostic/troubleshooting/benchmarking experience will tell you that if you are testing a particular premise, then all other aspects of the tests must remain the same to give a meaningful comparison. The people commenting that the test methodology is flawed are correct; it absolutely is and I just wanted to post my support for them. I'm not trying to convince you of anything though, I can already tell from the few replies you have posted on this topic that you are never going to concede the point.

People are funny.

 

[loki1944]

Wow so really not bad at all.

 

[poshflamingos]

I created an account purely to respond to your haughty tone and the rubbish you have posted. Windows does in fact use additional RAM if it is present - up to a point.

Really? You created an account purely to respond to me, and still completely failed to comprehend the conversation you were inserting yourself into?
The user I was responding to claimed that Windows uses available RAM as a disk cache. That is absolutely incorrect, Windows does not use RAM as a disk cache. Windows has prefetch, a feature preloads only executables and libraries. That's not a disk cache. Prefetch doesn't preload user documents, or game assets, for example (which a disk cache would be able to do).
I also said prefetch is limited to only a fraction of your total RAM, so the claim that Windows always uses extra RAM is false. Once it hits that limit, no more RAM is used for prefetch.

and the ONLY valid way of doing that is to keep the amount of RAM the same e.g. 1 x 16GB stick (single-channel, obviously) vs 2 x 8GB sticks (dual-channel).

You could not be more wrong.
1) No, you do not need to keep the amount the same for the test to be valid. If the games tested do not exceed 16 GB of RAM in usage, then the amount of RAM you have on your system is irrelevant to the test. You do not gain performance from having unused RAM sitting around in your system doing nothing.
2) You made this account just to respond to me, and yet you failed to understand what I already explained. A "1x16 GB vs 2x8 GB" test involving DDR5 would actually produce invalid results, due to the fact that 16 GB DDR5 sticks are dual-rank while 8 GB sticks are single-rank. If you test 1x16 GB vs 2x16 GB, and ensure none of the tests exceed 16 GB of usage, you are correctly ensuring only one variable is being tested, the memory channels. If you test 1x16 GB vs 2x8 GB, you now have two variables (memory rank and memory channels) and that ensures your results are invalid to compare the impact of memory channels, because they memory ranks are different. What you proposed makes the test significantly worse than what Steve did here, and you are completely unaware of that.

I have no idea what you do for a living but personally I've been working in IT Support (started as a technician, now a network manager) for 30 years in which time I have built/tested/diagnosed/upgraded thousands of systems, and as such I am supremely confident in my level of technical expertise.

LMAO
This is what we call an "appeal to authority" fallacy.
You are a random nobody on the internet, you word is worth nothing to other people here. And considering the fact that you didn't know what disk cache meant, you didn't know how Windows' prefetch actually works, and you had no clue what memory ranks in DDR5 are and how they work (and why they make a 1x16 vs 2x8 comparison invalid), I wouldn't trust your "level of technical expertise" the smallest bit. And you describing it as being "supremely confident" while literally writing a response to me that is objectively wrong is hilarious.

 

[gamerk2]

1) No, you do not need to keep the amount the same for the test to be valid. If the games tested do not exceed 16 GB of RAM in usage, then the amount of RAM you have on your system is irrelevant to the test. You do not gain performance from having unused RAM sitting around in your system doing nothing.

Would be true, if you disregard all the other things Windows does in the background that could be influencing the results. You also open up the risk significant paging could be going on in the background, which could influence (compress) the results. Remember even if the system doesn't use all the RAM, if it needs to grab a relatively large block at once and it can't be honored, paging will still occur, and that would absolutely affect the results. And it's invisible to the user (unless you disabled the pagefile, in which case you'd know because of the BSOD).

When testing, you test *one* thing at a time, simple as that. You tested *two*; saying "well, that doesn't make a difference", even if true, misses the point that you introduced a second variable into the test. Why you'd test like this instead of the much more obvious 1x32 versus 2x16 is beyond me.

 

[LimyG]

Reality is not a democracy. If you get together with a bunch of people who are wrong, it doesn't make you stop being wrong.

Democracy, Reality? No offence intended but I have no idea what you are on about??
No need to clarify.

Edit: Just read your other posts. Test aside, what's with the tone? Calm down man!

 

[meric]

Really? You created an account purely to respond to me, and still completely failed to comprehend the conversation you were inserting yourself into?
The user I was responding to claimed that Windows uses available RAM as a disk cache. That is absolutely incorrect, Windows does not use RAM as a disk cache. Windows has prefetch, a feature preloads only executables and libraries. That's not a disk cache. Prefetch doesn't preload user documents, or game assets, for example (which a disk cache would be able to do).
I also said prefetch is limited to only a fraction of your total RAM, so the claim that Windows always uses extra RAM is false. Once it hits that limit, no more RAM is used for prefetch.


You could not be more wrong.
1) No, you do not need to keep the amount the same for the test to be valid. If the games tested do not exceed 16 GB of RAM in usage, then the amount of RAM you have on your system is irrelevant to the test. You do not gain performance from having unused RAM sitting around in your system doing nothing.
2) You made this account just to respond to me, and yet you failed to understand what I already explained. A "1x16 GB vs 2x8 GB" test involving DDR5 would actually produce invalid results, due to the fact that 16 GB DDR5 sticks are dual-rank while 8 GB sticks are single-rank. If you test 1x16 GB vs 2x16 GB, and ensure none of the tests exceed 16 GB of usage, you are correctly ensuring only one variable is being tested, the memory channels. If you test 1x16 GB vs 2x8 GB, you now have two variables (memory rank and memory channels) and that ensures your results are invalid to compare the impact of memory channels, because they memory ranks are different. What you proposed makes the test significantly worse than what Steve did here, and you are completely unaware of that.


LMAO
This is what we call an "appeal to authority" fallacy.
You are a random nobody on the internet, you word is worth nothing to other people here. And considering the fact that you didn't know what disk cache meant, you didn't know how Windows' prefetch actually works, and you had no clue what memory ranks in DDR5 are and how they work (and why they make a 1x16 vs 2x8 comparison invalid), I wouldn't trust your "level of technical expertise" the smallest bit. And you describing it as being "supremely confident" while literally writing a response to me that is objectively wrong is hilarious.

Doesn't windows auto-allocate the paging file depending on the amount of ram installed on the system? That's if the user didn't set it up manually. So in theory more ram = less or no need to access to the paging file on the disk? I think the user you responded to wasn't talking about the prefetch but the good old disk caching system of windows. And that could be a valid point. Because if windows auto-sets the paging file, it can decide to not use a paging file when 32gb of memory is installed.

 

[ZedRM]

I think you misunderstood the question. Yeah, you have a single vs dual channel test, but at the same time you did 2x the RAM. Of course there will be gains there. The real test would be 1x16GB vs 2x8GB. Keep the amount of RAM the same, only test the channels. We already know that we're going to see a gain with 2x the RAM.

You have nailed it. I would further your idea, same speed and timings of RAM.

Such a poor test method. This is why Techspot is difficult to take seriously.

 

[poshflamingos]

Doesn't windows auto-allocate the paging file depending on the amount of ram installed on the system? That's if the user didn't set it up manually. So in theory more ram = less or no need to access to the paging file on the disk? I think the user you responded to wasn't talking about the prefetch but the good old disk caching system of windows. And that could be a valid point. Because if windows auto-sets the paging file, it can decide to not use a paging file when 32gb of memory is installed.

No, only the upper limit of the size of the paging file depends on your RAM capacity, but RAM capacity doesn't affect how much the paging file is actually used. On normal operation, the only thing that goes into the paging file is suspended system processes/services (for which the performance impact is null, they are suspended so they are doing nothing), and you're only using a small fraction fo it regardless of how big that upper limit is.
The paging file only gets used for active processes if you run out of physical RAM. Which is the whole point here, if you don't run out of physical RAM then there is zero capacity-related performance impact.

 

[poshflamingos]

Would be true, if you disregard all the other things Windows does in the background that could be influencing the results. You also open up the risk significant paging could be going on in the background, which could influence (compress) the results.

Neither of those things are real concerns. This doesn't exist, Windows isn't randomly, routinely running huge background processes that consume GBs of RAM, nor is "significant paging" happening during normal operation. Especially not so on a reviewer test bench, which won't have a bunch of consumer apps launching at startup.
You literally just made up an unrealistic scenario in your head that never happens, and tried to use that made up scenario to claim this test could be invalid.
That's as ridiculous as if I said that every test every website ever made is invalid, because a neutrino from the sun could have gone through the silicon in just the right way to flip a bit, resulting in a wrong result that invalidates the test.

if it needs to grab a relatively large block at once and it can't be honored

This is literally just gibberish thhat means nothing.

When testing, you test *one* thing at a time, simple as that. You tested *two*; saying "well, that doesn't make a difference", even if true, misses the point that you introduced a second variable into the test.

That's the whole point of contention. If it can be controlled, then it is not a second variable. If you ensure none of the tests exceed 16 GB, then capacity is not a factor in the test, so it's not a second variable, and the only thing affecting the performance is number of channels.
Your argument against this was 1) "what if Windows is suddenly taking large amounts of memory for background processes", which is not a thing that actually happens; and 2) the pagins stuff you wrote after, which was just incoherent nonsense. So no, none of those things invalidate that if you make sure you're not exceeding 16 GB of physical RAM usage, the test is completely valid.

 

[Raytrace3D]

Reality is not a democracy. If you get together with a bunch of people who are wrong, it doesn't make you stop being wrong.

Some argue that the games tested do not reach a single sticks memory capacity. That seems valid, but most people do not close / uninstall / remove programs that run in the background when using their computer to play games, like in this benchmark scenario. Most people are not technical. I argue that most users who are trying to save a buck probably also are not power users who know how to optimize Windows so that background tasks are minimized. Thus, I argue, having just 16GB's of RAM is a massive limitation for high-end games when running standard tasks. Shoot, even having Chrome running in the background or closed (but it's process running) can chew through 5GB's or more of RAM easily. As I type this message my system is using 33GB's of RAM having only a few programs open, mostly web browsers tabs. When my system first boots up, I'm running around 12GB's of RAM. Granted, I realize it's not practical to test this very scenario since everyone's experience is different, but it is practical to test a single stick of 16GB's or dual 8GB's or better, a single 32GB or a dual 16GB's of RAM. This way the user knows what dual channel actually will do when memory is not potentially exceeded, because the reality is most people's setups are likely to be impacted by background processes and memory constraints. Especially if this scenario is not on a newly fresh install of Windows (the most ideal scenario ever created).

 

[gamerk2]

That's the whole point of contention. If it can be controlled, then it is not a second variable. If you ensure none of the tests exceed 16 GB, then capacity is not a factor in the test, so it's not a second variable, and the only thing affecting the performance is number of channels.

You miss the point: Just because you have 16GB of physical RAM doesn't mean you have 16GB of "usable" RAM, since any requests for RAM need to be contiguous in memory. Throw in Windows loving to byte-align to word boundaries, and I can come up with several test programs that could "only" be using half the total system RAM yet fail on memory requests of just a few MB.

Simple test if you want to absolutely be sure this isn't happening: Disable the page file; any BSOD and your assumption fails.

But again, I ask "Why even bother to test this way in the first place"? This reeks of lazyness, and your aggressive defense isn't exactly helping your case.

 

[poshflamingos]

This is by far the most absurdly wrong comment anyone has made here.

You miss the point: Just because you have 16GB of physical RAM doesn't mean you have 16GB of "usable" RAM, since any requests for RAM need to be contiguous in memory.

Do you think we're still in the 80's? What you said has not been true since before Windows 95. Every OS uses virtual memory addresses, that can map arbitrarily to physical addresses and can be fragmented without any issues. This stopped being a concern literally 3 whole decades ago.

Throw in Windows loving to byte-align to word boundaries

The overhead from page alignment is so minuscule it's basically null. Maybe a few wasted KB of RAM mattered in the 80's, which is when it seems your understanding of RAM comes from, but when we're talking about GBs of RAM a few wasted KBs (maybe a few MBs at most) due to alignment is completely insignificant, and suggesting this could ever be the cause for lack of usable RAM is lunacy.

and I can come up with several test programs that could "only" be using half the total system RAM yet fail on memory requests of just a few MB.

Perhaps you could go out of your way to write programs that specifically target vulnerabilities/exploits in virtual memory addresses to cause it to fail. But software/game devs obviously aren't deliberately doing this, so this is a moot point for any professionally developed software, which is what's being tested here. Anyone who wants their software to actually work instead of deliberately trying to break memory allocation will not break memory allocation.

Simple test if you want to absolutely be sure this isn't happening: Disable the page file; any BSOD and your assumption fails.

Lots of people with very high RAM capacities disable the page file (even though this isn't really necessary nor a good idea), since this has been a recurring advice on tech forums over the decades for people who have an abundance of RAM. And no, that doesn't cause those people who do it to have constant BSODs.
I don't know where you got that ridiculous idea from. If disabling the page file meant guaranteed crashes, disabling it (or also not having a swap partition on Linux, for example) wouldn't be an option to begin with.

and your aggressive defense isn't exactly helping your case.

I'm not here to "defend" the test. I'm here to correct the people who are saying objectively wrong things because they don't understand how RAM works.

 

[poshflamingos]

Some argue that the games tested do not reach a single sticks memory capacity. That seems valid, but most people do not close / uninstall / remove programs that run in the background when using their computer to play games, like in this benchmark scenario. Most people are not technical. I argue that most users who are trying to save a buck probably also are not power users who know how to optimize Windows so that background tasks are minimized. Thus, I argue, having just 16GB's of RAM is a massive limitation for high-end games when running standard tasks.

Sure, you're not wrong. But that's not what this test is saying. The conclusion of the test isn't "there you go people, 16 GB is enough for everyone". The conclusion of this test is "this is how much performance you lose by using single-channel RAM instead of dual-channel RAM". Some people commented to say the test is invalid because there is also a difference in capacity between them (16 GB vs 32 GB). And I commented that the difference in capacity does not invalidade the test if the test never exceeded 16 GB to begin with.
The point isn't that common users wouldn't benefit from more than 16 GB. It's that, for the purposes of this test specifically, 16 GB does not invalidate its results.