4gb dual channel (2+2) vs 6gb single channel (4+2)

Hi. I recently bought an HP Sleekbook 14.
These are the specs:

Microprocessor 1.8 GHz Intel Core i3-3217U
Chipset Intel HM77 Express
Microprocessor Cache 3 MB L3 cache
Memory 2 GB 1600 MHz DDR3
Memory Slots 2 user accessible
Video Graphics NVIDIA GeForce GT 630M (1 GB DDR3 dedicated)
Display 14" diagonal HD BrightView LED-backlit Display (1366 x 768)
Hard Drive 500 GB SATA (5400 rpm)
Network Card Integrated 10/100 BASE-T Ethernet LAN (RJ-45 connector)
Wireless Connectivity 802.11b/g/n

I was told that there is still an unused memory slot and I want to add more memory. I want to use this laptop for gaming and photo and video editing. Which would give a better performance: installing a 4gb module (6gb total but in single channel) or installing a 2gb module (4gb total but in dual channel)?

 

6GB total would work in dual channel as well, called asymmetrical dual channel. Slight, but unnoticeable performance hit compared with full dual channel. I'd go that route. Considering you have a dedicated GPU though, the main performance advantage of dual channel would be for the GPU integrated in the CPU (Intel HD 4000), and you'd likely use the 630m for gaming. Although for the price of RAM, you can frequently find 2x4GB kits on sale for $45-50 for decent RAM. (unless the 2GB is soldered on the motherboard with only a single actual slot available, in that case get the single 4GB module)

 

I guess I'll go with 6gb then. The 8gb option though is kind of out of reach since prices in my country are quite high and I wouldn't know what to do with the remaining 2gb stick. Thanks for the response.

 

Great. Hope you enjoy your new laptop!

 

Likely to have somewhat better performance with 2 + 2

Also see: http://forum.notebookreview.com/lenovo/713260-q-thinkpad-ram-gpu.html

 

See this table I made in 2011. It's for Sandy Bridge, but the memory performance of Ivy Bridge is very similar.

John

 

Interesting. And the thread itself is interesting.

Thanks very much!

 

Here is the theory, do you believe it? Not really my concern. Does it work perfect in the real world? I don't really know.

Asymmetrical RAM Array, for most notebooks this means two SoDIMM's of different sizes. What can match up (pair) does and functions as "Dual Channel" in theory doubling the bandwidth vs one. Of course no memory controller gets even close to that, notebooks are far behind desktops on that. The remaining un-matched runs single channel.

2GB+2GB=4GB all runs "Dual Channel"
2GB+4GBGB=6GB 4GB "Dual Channel" 2GB "Single Channel"

Now pay attention, what is accessed/modified frequently is supposed to migrate to the matched/paired area of the RAM array. So as "Dual Channel" both should function the same as both have 4GB in "Dual Channel". But what happens when the needs exceed 4GB's? With the 4GB you do a Pagefile hit. With the 6GB you have an xtra 2GB b/4 you suffer the Pagefile penalty.

Does this work perfect? I doubt but the Pagefile hit is such a penalty that real world I suspect 6GB would function better than 4GB.

People like to say synthetic benchmarks are of little use and a bad representation of real world. While on many things such as that video card vs this video card or that drive vs this drive or CPU vs CPU I tend to disagree and find them useful. But benchmarks on the memory I do find to be problematic. Here is why. While I don't know all of them so can't be certain it seems. At least some benchmarks will say test 5GB but won't hit the Pagefile. So the 6GB gets a score based on 4GB in "Dual" and 1GB "Single. The 4GB will only get 4GB tested. Now say it tests 4GB. On the 6GB it might pull some of that 4GB from the 2GB un-matched once again giving the slanted non real world advantage to the 4GB.

The old thought that More RAM is better still holds true in my opinion. There is no real world penalty, also never confuse 2 SoDIMM's of different sizes with running a 1 SoDIMM, not the same at all in an Intel chipset. AMD I just don't know.

I am assuming we are just mixing size not speeds or latency that can complicate it a bit but in general more is usually still better.

Like I said does this all work out in the real world? Well my stopwatch broke and I don't have time to fix. Take it or leave it. Believe matched is better than more, all up to you.

 

I think you have said in words what my table, linked to above, has stated in numbers (and also shows, to some extent, the effect of RAM speed / latency).

I agree that the benefit of not using the pagefile is one that benchmarks find difficult to measure, so that is a further bonus of more RAM. However, the "more is better" rule breaks down if actual RAM usage is substantially less than the RAM provided. Some people talk of upgrading from 2 x 4GB to 12GB, in which case I ask them to check if they are actually using all 8GB before adding more.

John

 

Right. Remember RAM is dynamic and so is Windows. RAM is used as a storage pool too for opened apps, so unless you're actively utilizing all 6GB RAM at the same time, which is impossible given the bandwidth, you won't suffer a performance penalty. For Win7 and for most people 8GB is the sweet spot IMHO, both in price and performance. Beyond 8GB for most users they just won't gain any benefit. Only reason I have 32GB RAM is because (a) I can (b) RAM Disk is useful occasionally (c) more VM's.

 

talipho for you on your point I think even the others likely will agree that 6GB (1x2GB+1x4GB) is better than 4GB. HT says 8GB is the sweet spot and John says check your usage to see if you need. Getting up to HT's sweet spot I think more is better and Johns comments were I believe about larger amounts than 6GB. Only they can say for certain but at the 6GB point more is better even if that generalization I made does not hold true at greater quantities is agreeable.

I know it is but maybe others don't. Some might express the advantage for IGP's of matched. You have dGPU. So that reason out of the mix. And grabbing that last "fps" with a IGP for others I wonder if 14fps really matters vs 13fps?

If you had a 4GB stick and wanted to know if 4GB or 8GB to add? That would be a discussion. I would go 8GB to=10GB, but well not all would agree.

Good luck.

 

Perhaps your reading skills need to be improved?

For performance it's a close race and my take is the performance edge goes to 2x2GB over 6GB (1x2GB+1x4GB).

Meaning my read of that link of benchmarks of memory/S.B.cpu/Thinkpad-420s-laptop link that John Ratsey provided shows a small performance edge to 2x2GB over 6GB(1x2GB + 1x4GB).

 

Well that really hurts. I can think of three likely primary reasons I might lack comprehension.

From birth which might indicate I have some congenital or environmental issues I was born with. Or maybe growing up I faced serious either passive (neglect) or active (abuse) environmental circumstance. Or lastly and thirdly maybe I am suffering from some form of cognitive brain degeneration. Early onset Alzheimer's or even worse maybe Creutzfeldt–Jakob disease. All look bleak to me but thanks for noticing.

But if you consider your own comment about your read, you say John's chart shows a slight edge. I get that and I question the selection of what and where is the benchmark "reading from". I mean is it just reading stuff that has no meaning? I think it is clear it did not open up anything so as to move into memory to then as a real app would do make use of it. Say make requests from the CPU for any real function or purpose. That is my comment on the bandwidth. Now consider on the two "Vantage" benches that I neither endorse or consider a real world test. Bottom two on the chart are for any real purpose the same. They are so close they count as the same.

Other Song we are discussing something that well is really not the most critical thing in the world. We disagree so be it. You will not sway me. If I won't sway you. so be it. I think more in this situation would be better and I think the fact that my thought offers the exact same amount of paired/dual channel +2GB xtra is the way to go. If OP picked your way of 2 equal matched I doubt that there would be any performance issue real world. It is a bit esoteric this discussion and all. I mean I guess in many ways it absolutely does not matter as far as perceivability be it pagefile hit or not. HT mentioned the OS being designed to "manage" "things/tasks" and he is spot on.

I know there is not really a penalty when you have a 2/3 paired and 1/3 not, ratio of RAM when dealing with these non extreme amounts of RAM. At the same time you could legitimately tell me that real world there might not be advantage to my 6GB? It would be valid in most real world situations. The CPU request (loading a program) knows what might be needed and behind the scenes moves data into RAM ahead of time, kicks other things out. I am not going to get into if it is an algorithm or not but it continues doing that. Makes moves ahead of time so you never see or feel. Not saying my 6GB is any different works the same. I might have better luck mostly if something was lucky enough to stay in RAM on the 2GB extra but that really is not worth concern or maybe it is.

Real world it does not really matter. But all you guys who think unmatched pairs decrease performance real world are wrong. This discussion has been going on for years. The good old days like pairs must be the same and matched? You had to buy Dual channel RAM? That was the best because there is no such thing as dual channel RAM in and of itself. It is a function of the MoBo. I hope that does not start a "row" with my last days of cognition.

Spend $5 extra bucks and get 6GB or save the money and 4GB. It really does not matter that much.

 

I agree that 6GB is likely to give overall performance better than 4GB in spite of the memory performance being slightly slower due to reduced likelihood of needing to use the relatively slow pagefile. Also, using any combination of two modules is better than one.

John

 

But the Windows pagefile isn't slow when one uses a SSD, and not a HDD.

Do you have any idea how using a SSD might change your memory test results???

 

While I agree that an SSD is faster than an HDD (but maybe only 5 to 10 times faster for reading or writing a block of data - a lot of the SSD advantage comes from the near-instant access time), an SSD is much, slower than RAM access (RAM = 10 to 20 GB/s while SSD is 250 to 500 MB/s. In addition, hit any storage device with a big data read that triggers a corresponding write to the page file and it won't perform at the optimum speeds. (The Sandisk U100 SSD used by Samsung in the Series 9 is a particularly bad offender in this respect - I've seen the computer effectively freeze for a few seconds).

The results in my table were measured on a computer with an SSD (Intel 320). The SSD boosts the PC Mark results compared to the same computer with an HDD. You can see this if you look at the performance test results in my T420review.

Also, here's a very recent pertinent comment about the effect of RAM configuration and size.

John

 

Thank you for the thoughtful response.

Interesting review of the ThinkPad T420s.