Is dual channel ram a joke ?

...for average/ non-enthustiast users ???

I mean, i have right now a pretty powerfull lap with dedicated gpu (asus N550JV) which idles around 10W which wifi on and full brightness and taking off a ram stick can lower idle power consumption by another 0,5-1,5 W. So the impact on battery life is noticeable at this stage.

So question is :

Is there any scenarios where dual chanel ram really offers substantial perf gain over single channel other than for igp ??????

I think one 8GB stick of 1,35V is enough when your idle power consumption is low and you care about battery life.

 

If maximum battery life is your main concern, I would agree.

If maximum performance/responsiveness per watt is your goal: Dual Channel RAM is a must.

 

Any proof/link to back up this statement ?

 

Here's a quick google search:

See:
Is Your Gaming Laptop's RAM Slowing It Down? | Maximum PC



As usual, the link above shows that 'it depends' on your specific workflow/program/game.


Over the years I have undervolted, throttled, dimmed the display and used other tricks to extend the battery life of my notebooks.

However, I have always undone any 'tweaks' that crippled the performance of the system even at the expense of battery life. A stuttering/pausing platform is nothing to be proud of when you get even half an hour or more of (unresponsive) extended use in return.


If the system is just going to be on/idle - I can see why removing a Sodimm might be tempting.

If the system is being used (enough) though - the responsiveness of the dual channel system in question will allow you to do more (even with less battery run time) than the same single channel system with potentially more (at idle) run time.


The linked testing is a little quirky:

They tested with a 32bit O/S.

They didn't test for battery life.

They didn't test for anything other than gaming.



With my hands on experience with different systems:

I always recommend 2x2GB configurations to clients - even those with 32bit O/S's - dual channel makes that much of a difference (and I've had to show a few that 2x4G is better than 1x8GB first hand - they marveled how obvious their setup was better with the same amount of RAM, but in dual channel mode).

Navigating the Windows O/S is smoother too. Not to mention web browsing, opening programs and PDF 'portfolio's' and anything related to a database (LR5, accounting software, etc.).


Yeah; responsiveness is a high priority for me (and I'm keenly aware of the lack of that 'snap') - I would be hard pressed to give it up even for two hours longer run time - let alone the few minutes I have seen in my systems.

 

Intgrated graphics depends greatly on dual channel RAM. Video/Audio encoding is also sensitive to dual channel RAM and RAM speed. Otherwise daily tasks and even gaming with a dedicated GPU it makes nary a difference.

 

..um, well. What essentially happens (and I know this isn't completely true, so don't sue me), is that when you have a dual-channel option, you can effectively read from two places in the ram-areas at the same time. And since industry standard ram (outside certain RAMBUS solutions) only ever has one pipe, the idea was to let the memory bus access individual ram-chips at the same preparation run.

You would want to do that because in certain situations that would mean you could prepare and read twice as fast to complete a read-operation. Or, more commonly, to read to one module, while writing to another (which would typically prepare for effective reads later). So.. great, right? But the drawback being that you're not really going to have all that many situations where a program writes to ram while preparing for a dual channel read. And in the massive majority of cases, you're simply not going to see any improvement at all. On average, we're perhaps talking about 10%.

But. Some programs will exploit it very effectively. And drivers that are written well can exploit dual-channel reads for operations that are typically meant to run independently on specific, small, memory areas. Since the program can prepare two blocks instead of just one each read. APUs and igps are a typical example here, where shader operations on a graphics card are pretty much completely parallelizable.. (or whatever it's called in English).

And you actually see that while it won't cut the running time of the program all that much on average, leaving reads and write queues with quicker response makes the system feel more snappy. So it's.. yes. A bit of a joke, but not completely.

 

An article on integrated gpu performance lol. Pointless when u have a dgpu.

You all talk about system responsiveness as it is even really noticeable when we talk about big stick of ram. I don't think so. We're not not speaking about one stick of 2GB that is easily overloaded in multi-tasking but 8GB here, which is enough for about everything.

High frequency ram and multi-channel have always been kind of elitist stuff for enthusiast and overclocking when ram frequency was linked to bus speed, if not for overcloking, it was giving just 5% of better perfs in best case scenarios.

HW do you think it really matter in video encoding ? I will dig into that.

[Discussion] RAM: Single vs Dual Channel speed benchmarks (1x8GB vs 2x8GB) : buildapc

As always kallogan is friggin right

There is even cases where single channel give better perfs lol

 

Mm. It's going to be rare to see actual drop in performance, I think. Since we're not talking about parallelism on larger areas than just one read operation on the low level anyway. And without threading, which might explain the numbers in that test, you're only ever going to see dual reads and writes. Which could marginally reduce the response time in preparation sequences and under sustained loads when nothing else is limiting speed (see the min fps column).

 

The largest benefit is for IGP. Not sure what you're going for Kallogan? It has been known for a long time that dual channel will help performance 10-15% at best with specialized apps or specific tasks, usually to do with just number crunching like encoding, compiling, etc.

Single stick won't affect battery life much either, especially at idle where the sticks consume less than 1W. I'll take my 16GB at dual channel and spend an extra 5 minutes less of battery for the more RAM that is more useful to me. If 8GB is enough for you, then sure thing, go for a single stick.

Fast RAM, like 2133MHz+ can make a noticeable difference in system responsiveness over DDR3 1600, whether dual or single channel.

 

Sorry if i'm a litlle provocative, it's just i'm tired of hearing bullsh*ts and marketing scams about ram. Dual channel existed now for years, not to mention triple channel failure, and still didn't demonstrate any substantial improvements if not for marginal exceptions.

I myself tried 4GB vs 2x4GB in many games and it didn't bring the slightest improvement framerate wise and barely 30 pts in 3dmark.

Now i'm aware that we'll enter in a serious igp era with broadwell and Kaveri where ram will probably matters more. Well i'm not even sure since they tend to put dedicated ram on die now. If ram is cheap no problem but if not, prepare to pay for the high frequencies ram race to boost ur igp

And 5-10% on 5 hours battery life is 15-30 minutes more for my case. So for 10 hours...

 

Introduction

Test Results: Single Vs. Dual Channel RAM - Parallel Processing, Part 2: RAM and HDD

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/11390-intel-core-i7-nehalem-920-940-965-xe-processor-review-11.html

 

All is said. Marginal perfs gain. Pointless except for the hardcore bench junkie. Thx for the link.

 

For IGP there is significant improvement, fwiw here's some testing I did a while back: http://forum.notebookreview.com/har...0g-igp-tested-various-ram-configurations.html

And I did test with 1 stick DDR3 RAM and also two sticks with a dedicated card with no difference in performance: http://forum.notebookreview.com/gam...amd-radeon-hd-6750m-benchmarking-results.html

One of the mods here did something similar with an Intel setup and had similar results.

So as it is, minor to no benefit for daily tasks, only IGP is where it matters.

 

Single vs dual channel doesn't mean much with a dedicated GPU. Ever since DDR2, I have never been able to tell the difference between single, dual, or asynchronous RAM. Well, that's not true. I had a HTPC that used a Nvidia IGP that I switched from single to dual channel DDR3 to get better framerates in games, but that was the only time.

And as usual, Tiller is making claims that are refuted by both benchmarks and personal experience.

 

The difference between single and dual channel for the most part is a tiny improvement in performance that can only be detected in benchmarks versus a tiny improvement in battery life that again can only be detected in benchmarks.

Why are we even having this discussion?

 

To make you talk and because it's necessary to make things clear once for all.

 

No on die memory for igp that I'm aware of, even iris pro has on package memory (and only 128MB at that, so basically a buffer).

Back on topic, it depends on your usage, but if it's just basic usage and you're not using an igp, single vs dual channel shouldn't matter so long as you have enough ram.

3d modeling, photo editing, video editing and APUs are what mostly see benefit of dual channel ram.

In my case I have an A8-4500m in my laptop and just by moving from single channel to dual channel I was able to increase my gaming performance by about 50% (in benchmarks as well as fans), for other applications and general responsiveness, I noticed a small improvement some areas such as application loading time and benchmarks were around 7% better in tests where memory was a factor.

Sent from my SGH-I717D using Tapatalk 4

 

Well, so much misinformation here.

System performance (whether with igpu or dgpu or both) depends on the bandwidth of the platform's RAM.

Period.

Granted, some programs/applications/workflows will be more sensitive to memory bandwidth than others - hence the 'it doesn't matter for most people' stance.

But what I answered is still true:

If people here learn the basics of a computer system, they won't argue that.

CPU+RAM=WORK


Have a higher/lower performing cpu? The work performed will track. Same for the RAM. That is the definition of how a computer works.


In the link I provided, the results were scoffed at because of the use of an igpu. What?

When I continued with my experiences, including the observations others have made (my clients) that was discounted entirely. What?


Sorry, but you can't have it both ways. Either some numbers, 'scores' will point to the conclusion I've reached - or else more subjective results need to be taken into consideration. Not because some of these improvements are only subjective - but because they simply haven't been seriously looked at by the tech press on the web.


HTWingNut, your testing shows how dual channel RAM at MUCH lower frequency/performance matches higher frequency single channel RAM. Yet your conclusions state that single channel vs. dual channel performance improvements are moot.

In the link I provided, those improvements from higher frequency RAM is in the range of 10% (from a 30% improvement in memory bandwidth).

Based on the above, I disagree with your conclusions on the single vs. dual channel benefits.


Sanjiro has it correct when he states that he saw improvements even in non-gaming workflows.

That is what I see exclusively (I don't game).

And even if we go by his estimations (7%) that is still huge to me (obviously, we have a lot of people here who are insensitive to an O/S's responsiveness).


Kallogan, what your original post asked and what your latest conclusions allude to are not even in the same thought process.

I did answer your question, but you simply don't want to accept it.

Just because in your case you saw no difference in 'scores' doesn't mean the other benefits don't exist for others.


This is not a scam or bs marketing - the reason we're seeing real world results (mostly) in the single digit improvement zone is that memory bandwidth is still orders of magnitude below processor bandwidth (not just the L1 cache bandwidth) - so even doubling the ram's contribution is still leaving the cpu starved for more. And implementing dual channel (or three or four) is not an easy task at the speeds the modules run at.

Not only is the physical implementation much harder - the modules themselves are (mostly) unknowingly tweaked to run at all.

The MB, BIOS and the O/S and programs also need to be optimized to leverage the benefits of multi channel memory.


This story shows it best (in a quick search):

See:
DDR3 Memory Performance Analysis on Intel X79 - Intel X79 Memory Performance - Legit Reviews



I agree that the day to day improvement is small. But that is not what you originally asked; to me it is significant enough that any battery savings (up to 2 extra hours at idle, yay) is not worth the loss of the system's responsiveness during the whole time I am using it.


Now, have we put this 'bs' to rest?

 

Dual channel definitely increase performance, but everyone's perception is difference. Some may not feel the improvement at all( not a rare scenario in this thread), and imo it is a perfect valid stand point.

 

We're talking some serious special cases if you consistently see no improvement at any time, though..

That being said (thinking about the fps-thing in the other post). You could imagine a program flow where running time is not dependent on memory operation in main memory. Or that outside preparation and loading sequences, no task will be bound to submits or resubmits in ram. And that would basically be more or less all xbox and xbox360 games ever made with that hardware in mind, for example

But the "potentially twice the amount of memory operations" thing isn't made up. Even if the actual payoff is comparatively smaller nowadays compared to when ram was made of wood-shavings, you can and will absolutely shave off the response time during sustained writes, or for example see fewer forced waits for threaded background tasks.

 

Just tried one stick of Crucial 8GB 1600mhz 1,35V vs 2x4GB Kingston 1600mhz 1,35V, and Windows index, which is not a reliable reference i know, gave a 7.9 score for both.

 

Just curious. It is well known that slower CPU is then less useful it is using theoretical RAM speed. While more Mhz it has bigger bench scores in RAM it gets. AKA quad core I7 Ivy can get lets say 20GB/sec of 25GB/sec theoretical speed of 1600MHz RAM dual channel. While slower ULV dual core I7 may get lets say 15-16GB/sec. And even lower in single channel.
I would be curious to see performance difference in ULV CPUs and I3. Not in games but in apps load time and boot time.

 

As if ram was the limiting factor when loading apps and all, it's the HDD or SSD, and maybe the cpu power at some point (compressed files etc) but certainly not ram.

 

Im curious:
How many notebooks today only use 1 stick of RAM?

Also, I don`t think people give a damn about they should use 1 * 8GB or 2*4GB but its up to most OEMs to decide. And I`m pretty sure that over 90% of OEMs use two sticks, that means dual channel.

You won`t save much of those green $ by going 8GB instead of 2*4GB anyway since RAM is cheap.

 

OEMs lol. They use 2x4 because it's cheaper than 1x8 and because the common thinking is that dual channel perform better. As James D said, ram has more bandwidth than any of the other components even in single channel so i don't care about what oems do.

Just sayin' that's all, feel free to do what u like bro.

 

TBH dual vs. single is over played. Most systems are fine with just single channel. IGP systems depend more on dual channel where the IGP is not on die. The older Intel IGP's definitely took advantage. The newer ones can benefit from dual channel as well but just not as much as the older off die.

As mentioned it is cheaper for dual compared to single stick configurations. This is why OEM's tend to go that route. Also why they tend to go with 6GB as a 4GB+2GB as well rather than 8GB, saves cost but still gets you above the stock 4GB.

Now if you are just running SuperPI and that is your primary use, then yes be sure to run dual channel and as much of it as you can. If you are a real world user don't worry about it so much. Yes the faster the ram the snappier it can fell once things are loaded to ram, but here lies the question. If an action originally happens before you eye opens when you blink it then how can you perceive it being faster? In other words at what point is fast just fast enough?

 

Fast is never fast enough. Not in computers. Never has, never will be.


Our perceptions on what is 'good enough' changes slowly over time - but try going back to an older, 'classic/not upgraded' machine for your current workflow - it feels like it crawls and seems like I can see the gears turning in the cpu to spit out the answers that I am getting older while I'm waiting for them...


I get it. Some people don't see/feel a difference. I have some clients like that.

The people that do feel sub second (and even sub millisecond) improvements should not have to defend their decisions either though - no matter what the 'scores' suggest.


And the reason OEM's use 2x sticks of ram when they can: it's simply cheaper for them at that time (not the better performance option).

 

Agreed 2x stick's cheaper. Agreed with not defending purchasing to make it the absolute fastest, I do it myself. My point is for most fast is fast enough, it does not need to be sub millisecond fast.

So for most casual users fast is fast enough. This to the point the focus is no longer sole speed but size and power consumption for a lot of people. This is the OP's point, he seems more interested in power consumption than the super fast processing. Which is fine too but this is why there are so many options out there. Just one configuration will not fit everyone and their usage...................

 

...thank you all for the unusually sound philosophy lecture.

But the reason why you would focus on IO response on the fastest "user level" in the first place is program response. To get programs to feel snappier, or have fewer stalls. And without a strategy for handling "user level" interrupts differently (having UI running independently, asynchronously from transfer operations, along with strict rules for how programs are loaded.. which there isn't on your typical OS.. well, Windows, anyway), reducing io-bound stalls in the program space is the most natural way to make a typical x86 computer feel more responsive. ..will be that way until we ditch the crap for scheduled io and programmable instruction sets, probably.

 

How I see today's 'this is fast enough argument' is the same I saw for netbook around 2010...

When I finally bought one myself to see what the fuss was about, I fell in love with the weight, the size, the good enough keyboard/mouse/touchpad but the so called 'good enough' performance chant that was trying to be rammed down our throats was laughable.

Even with an overclockable platform (MSI Wind) the hit to my productivity was atrocious - the little netbook was returned within 6 hours.

I know people think we're far from that place right now (and the 'scores' would agree with them) - but I don't see the difference at all.


Performance is all computers are about (even if all we do is facebook/surf) - compromising on the cpu power is like going back to 1971. Or, not caring that we're dying faster than the cpu is working. Yeah - a little dramatic, but it isn't far too from the truth either.

 

Why is this even an argument? Why do people care? If 1 stick works for your Kallogen, then run with it. We don't care. Dual channel RAM is difficult to measure its perceptible improvement in performance but I think nipsen and tiller hit it on the head. To put it bluntly a "snappier" system. It's difficult to measure and show any real world value, but over time if you've used the system as tiller noted, it's a phenomenon I've noticed as well, is going back to the older tech seems slow. It's like the new tech doesn't necessarily wow you, but it isn't until you go back to the older tech that you appreciate the newer. Dual channel and faster RAM is the same. Try going back to 1066MHz single channel Core 2 Quad system compared with a quad core i7 with 1866MHz dual channel RAM. The entire package difference is phenomenal. It's the combination of the parts that matter, not necssarily the single tech performance improvement.

 

While the performance increase can be described as imperceptible, it must be known that the battery life improvement from losing the extra RAM stick is also imperceptible since RAM isn't a big energy consumer, you are talking a matter of maybe 10-15 minutes here. On Ivy bridge systems, dual vs single channel from a power perspective is also very minimal since the CPU can power gate the whole RAM bus link when it isn't in use.

However, addressing the performance issue, the main advantage of Dual channel is that you can use higher performance, low density modules without compromising capacity. A single 8gb module loses to 2x4gb modules when it comes to latency in addition to having half the bandwidth, this is mostly the factor which some people notice Dual channel configs as being snappier.

 

My table here is getting a little old but it is still valid to show the performance difference between single and dual channel RAM. (Note there's an error in the ram config in the 2nd last row).

Regarding power consumption, I believe that one module will draw about 2W when under full load (actively reading / writing) but the power consumption will be much less if the only activity is to refresh the RAM contents. I would also hope the system is clever enough to only refresh RAM that is in use.

John

 

yeah dual is probably a little snappier in theory but for the recall, we're talking about nano-seconds latencies. What can i say, seems like you all have some super spider senses. Good for you.

 

The difference between single and dual channel RAM will be felt more if you are using software that is using a lot a bandwidth. The integrated GPU will probably be one of the easiest ways of testing the difference as it will be more bandwidth limited than most things.

 

You can also escape band-width heavy critical sections that would otherwise stall program response. Say, when a program suddenly requires reads that take precedence, and stops other programs from performing IO. And you can measure this in the sense that in one case you have half-second stutter, or no response, while doing some often fairly light multitasking. Against not having those gaps at all.

(I mean, seriously, it's very easy to create a theoretical example, and even easier to find a practical case where loading times, or response while indexing, etc., is higher with dual-channel ram..)

 

I've been a computer nerd for a long time, and I have never heard this nonsense before. I was trawling through google for an hour trying to find others who made these claims, and there have only been a handful in the 11 or so years that we have had dual channel memory. And nobody ever agreed with them. There are countless articles and discussions of single vs dual channel RAM, and system responsiveness is simply not one of the differences. If you want more system responsiveness, get more RAM, an SSD, and/or a faster processor

Now I will explain why "dual channel RAM = more snappiness" is complete BS. The CPU needs numbers to crunch, and if it doesn't have it on the L1, L2, and if it has it, L3 cache, it ask for it from the system memory. From the moment the request is made to the time the CPU starts working on that data is (with CAS9 DDR3-1333) only 13 nanoseconds. This doesn't change between dual and single channel RAM.

So the CPU might start working on something within 13 nanoseconds of asking for it from RAM, but it might work at a slower pace as it waits more for the rest of the data that it needs to really get cranking. But how much more time does it need to wait? Is it the half second that nipsen claims can plague his computer, or can tilleroftheearth, with his superhuman perception of "sub-millisecond" times really feel a difference? The answer is neither. Single channel CAS9 DDR3-1333 will get a core to start work in 13 nanoseconds. The entire L1 cache of all four cores of a Haswell quad-core that have already been doing work since 13 nanoseconds can be filled from nothing in less than 24 microseconds. Finally, 120 mircoseconds after requesting data from the RAM, the entire L1 and L2 cache of all cores of a Haswell quad-core can be filled with completely new data, although for almost all that time, the CPU has already been doing work.

So there you have it. If your computer is lagging by half a second when you are using single channel RAM, but doesn't do this at all with dual channel RAM, your computer has serious problems. If you are like tilleroftheearth and are an alien, android, and/or robot that can detect lag at a threshold below one thousandth of a second, you are in luck because even cheap single channel DDR3-1333 will be fast enough for you to not perceive any lag.

 

Laptops with single channel memory controller, I call them crippled laptops. I could not buy a laptop without Dual Channel memory support. In desktop front Core i7 4960X has quad channel memory controller .

 

I'll admit it's rarely going to make any difference in most benchmarks. Since most "heavy" lifting nowadays is and almost inherently will be cpu-bound. But you can create situations very easily on a system when running threaded tasks where you're getting IO starvation for pretty simple things.

Say you're running a java compilation in the background, and listening to music you're streaming from somewhere. Since you're running these two tasks concurrently, w,What's really happening now is that every now and again the OS will perform a context shift. And at that point you can end up filling the queue requests, and both tasks will have to wait until program control is returned. That's when twice the bandwidth counts.

You can see the same effect in one of the tables in one of the first posts in the thread, with the min/max fps numbers in a game. On dual-channel the lower fps-count is higher, but the max fps count actually suffers a tiny little bit, most likely because of fragmentation or some writing or preparation quirk. But that's.. kind of a very good example of the benefit/disadvantage proposal with dual-channel.

I'm not going to say the same for quad-channel and triple-channel. But for dual-channel modes, you're shaving off a fairly typical situation that happens in practice on.. any system, really. Anything from running a compression on a mic stream.. a skype stream, and over to just having a wrapper for a gamepad running during a game, etc., can cause these kinds of crunches very easily. So... you know.. Not essential. But there are situations where you can see the difference very fast. (Without running around with a hair-hygrometer style wishing rod, or something like that.)

 

So does anyone think it is worth it to pay $90 to add another 8GB stick of RAM to a Sager NP7378 (where I already have a single stick of 8GB RAM)? I would of course double the amount of RAM, as well as taking advantage of dual channel. I don't do video editing. The most taxing thing I plan to do with it is StarCraft 2.

 

No, you won't see any difference.

 

It will only impact games like battlefield 4 which are highly threaded so need the bandwidth to feed the cores.

 

Thanks I decided to just stick with one stick of 8GB for now. Hopefully SC2 is not highly threaded.

 

Nope. Fast dual core is all you'll really need. More cores is almost useless for SC2.