The Most Common DDR DRAM Myths Debunked

References
By
published

There Are No Performance Gains With DRAM Faster Than 1600 MT/s

The answer to this depends on many factors. It is completely false if you are using a CPU's or APU’s on-board graphics. They use your system DRAM—and the faster, the better!

Most DRAM benchmarks measure read, write and copy performance. Many gaming benchmarks demonstrate a 3 to 5 FPS gain on DRAM between 1600 and 2133 MT/s. That’s because, in most games, DRAM is used primarily as a pipeline to feed information to the GPU and as a holding area for frequently retrieved data. The fact remains that it can increase FPS a little. Because the price of DRAM between 1600 and 2133 MT/s is only narrowly differentiated, it can sometimes still make sense to get the higher data rate DRAM.

Furthermore, the file compression program WinRAR pulls the data into DRAM and compresses it to DRAM before writing it back to the drive. Benchmarks using WinRAR can show a 25-percent gain going from 1600 MT/s to 2400. And there are plenty of other memory-intensive applications: editing video, working with images, CAD, running VMs and so on. While some of the gains might be small, the little snippets of savings add up if you use those applications.

If you are doing one task at a time—writing a memo, then browsing a Web page and then watching a video—you don’t really need faster DRAM. If you multitask, however—for instance, if you have a bunch of browser tabs open while you're working on a good-size spreadsheet, or if you're running video in a window, are working with images and maybe running a virus or malware scan in the background—then the faster memory can be far more beneficial.

You can test this out by running some applications like these with 1600 MT/s DRAM and then something faster. Once you’ve loaded up your system with several programs, use something like the SiSoftware Sandra system benchmark, and then use WinRAR on a large file. While it runs, hop around through your open Windows, and then check your scores in Sandra and the elapsed time for the WinRAR.

Jim Reece
98 Comments Comment from the forums
  • damric
    Well that was a good read. Thanks Tradesman!
    Reply
  • Nuckles_56
    Thanks for the article, I've been impatiently waiting for it since the first one
    Reply
  • JackNaylorPE
    Some if these could use a bit more detail.

    "All DDR3 is the same" ... Just wanted to mention that in close on 25 years managing and participating in web forums, can't say I have ever seen this one put forth for either DDR3 or anything else. There are numerous differences, some matter not at all, some matter a little, some can matter a lot. It's crapshoot, but the odds are pretty good if you don't have a lot or large differences.

    "Mixed DRAM Runs At The Speed (Or Timings) Of The Slowest DIMM" - I think this is just poor wording of the intent when this one is used. More correctly, it's much like video cards .... one card is advertised and guaranteed to run at one speed, the other a little bit faster. Like RAM, in SLI / CF you have a pretty good likelihood of setting them up to run at the lower of the two speeds. You have a lesser change of getting them both to run at the higher speed, tjo it's worth trying; that slower advertised card may have some headroom that was previously untapped

    You also have the issue that 2 cards put more of a strain on the PSU and when that extra load is present, voltage stability suffers. Your faster card, with more voltage variation may not be able to handle the same OC that you had before the 2nd card's load was added.

    Same w/ RAM ... one would presume that if a set of RAM was capable of a higher speed, it would have been binned and sold at the higher speed. As vendors respond to the variability of supply and demand, we may see better quality sets sold at lower rated speeds, which we can attest to by reading how many reviews in which the tester was able to get significantly higher speeds. So when someone says "will run", I have always taken it to mean "the likelihood is that it will be able to run at the less demanding of the two speeds / timings. As always, no guarantee. As it is, your likelihood of getting 2 1600 sets to work together is much higher than getting two 2400 sets to do the same. Success will be better at 1600 / 1600 and 1333 / 1600 than it will be at 2400 / 24500 or 2400 / 2666

    I haven't been asked to do many upgrades since Sandy Bridge days (when some folks were still buying 2 x 4GB) but back then at 1333 / 1600 and 1600 / 1600 matchups, our success rate was well over 90%. And those difficulties came usually resulted from having different module OEMs. In the last 3 years, I'd say 9/10 builds started with 16GB and those who chose 8GB are still content, so don't get the opportunity any more.

    "Just add more DRAM" .... I'd add that the most common cause if mismatch results from the maturation process of the production lines.... When DDR3 first broke 1333 was the most commonly purchased speed, as productions lines matured and yields improved this grew to 1600 and of late we have seen great yields on 1866, 2133 and now even 2400. As lines mature and yields improve, required voltage also comes down.

    Initially, a vendor will usually buy their lower speed modules from one vendor and then sign on with a higher quality vendor to fill their needs for higher speed RAM. Over time, the lower price vendor's yields may be sufficient to supply their demand for the higher speeds. So we wind up with a situation where a vendor's product from 2012 will use one brand of module and one from 2015 will be from another. That presents quite a challenge for compatibility.

    This can be particularly frustrating as one version from January 2012 and one from November 2014 might work fine and then one from November 2014 and January 2015 might have no chance in "H E double sticks" even tho every one of them has the exact same model number. Sometimes you can notice this by a slight change in timings. Several vendors offered 2400 at one time at 10-12-12-28, nor those same models are 10-12-12-30 or 10-12-12-31.

    "It’s Cheaper To Buy Two Sets Of DIMMs Than Larger, More Expensive Sets" That's not a myth, at least not as stated.... they are cheaper. The "myth" as stated doesn't say anything about working :). They may not run together at advertised speed but they are still cheaper.

    "There Are Only A Few DIMM Manufacturers" .... given the limited 2 sentence attention this subject got, the topic should have been eliminated.
    Reply
  • djsvetljo
    Wait a second, I thought boards like Z97 have their own Memory Controller, separated from the CPU and 2400Mhz RAM is usually not a problem with Z97, am I right?
    Reply
  • TechyInAZ
    As always, thank you Tradesman! Such a big help.

    Wait a second, I thought boards like Z97 have their own Memory Controller, separated from the CPU and 2400Mhz RAM is usually not a problem with Z97, am I right?

    Nope, CPUs have the MC in the CPU itself. It's faster since it's directly in the CPU which also saves energy compared to the old way of storing it on the mobo.

    I personally run 2400mhz on a Z97-A with a i5 4690K and I've had no issues.
    Reply
  • Dark Falz
    There were instances with past chipsets where 4 DIMMs would be faster than two, due to the way the chipset interleaved memory or leveraged reads. This was true of my Pentium 4 system (yes, I realise how old it is). Probably stopped with the IMC. I picked up a second 8 GB kit from eBay a few years after putting my IB together and lucked out, despite a few months difference in mfg date they work perfectly with the existing kit and appear to the same package/chips. They run at 1.50v too when they are rated at 1.65v. TBH I haven't yet seen any benefit from 16 GB vs 8 GB for anything I run, but I mostly just run games and Chrome. I suppose there's 8 GB more for Windows caching but that's less important with SSDs and modern streaming engines.
    Reply
  • Tradesman1
    16227476 said:
    Well that was a good read. Thanks Tradesman!

    ____________________

    Appreciate it ;)

    Reply
  • Tradesman1
    16227672 said:
    Thanks for the article, I've been impatiently waiting for it since the first one
    _____________________________

    Thanx, have had a number of PMs and emails about when this would be published, it's been been done and waiting ;) I will say, the staff at Toms is small and they are are no doubt overwhelmed with the number of pieces that they publish. Hopefully it will be of help to many

    T
    Reply
  • Tradesman1
    JackNaylorPE

    Comment - “Some if these could use a bit more detail.”

    Always and forever true. In this particular piece, many items had more detail, examples and explanations as written. However I have no say as what they do with the piece once accepted. Originally it was one piece, with a working title of “DDR3 – FAQs and Fiction”, (have no idea where they came up with adding “and Troubleshooting Guide”, a guide to troubleshooting would be a step by step thing).

    Comment - “All DDR3 is the same”

    In the first piece introduction I stated “purpose of this article is to address the most commonly asked questions we hear, and to debunk some of the myths” - The “all DDR3 is the same” is a comment seen daily in the memory area of the forums, and as with the other is a myth, as I mention, this subject alone could be the basis for an article (as could others)
    Comment – “Mixed DRAM Runs At The Speed (Or Timings) Of The Slowest DIMM"
    Again this is a common misconception that is seen daily, and thus was included and explained.

    Comment - "Just add more DRAM"

    First, with the intro of DDR3, back with the 775 and 1366 mobos 1066 was the prominent data rate being sold, 1333 and 1600 were considered the ‘enthusiast’ data rates of choice. The 1366 CPUs were rate 800-1066. The early 1156 Pentiums and Celeron CPUs were rated 1066 and then the Clarkesdale and Lynfield i3-i5-i7 CPUs were rated to 1333. (Also keep in mind, the original JEDEC specs for DDR3 only went up to 1600 which was the max data rate).

    This Item in particular was much larger and had an example of of a vendor making a model of a chip, binning it to different levels, selling them to different manufacturers who further binned them, etc, etc.. That part wart was sliced and diced ;) Here you talk of chips produced years apart and seem to miss the point that you can sticks of DRAM right off an assembly line and they might play nice, they might not. This is why manufacturers test DRAM that goes into a package

    Comment – “It’s Cheaper To Buy Two Sets Of DIMMs Than Larger, More Expensive Sets"

    Keep in mind that this article was aimed at statements often see/heard coming from others giving ‘advice’ propagated by ‘experts’ and stated as a fact to those who are looking for 4 sticks for a single rig.

    Originally I phrased it as “Just buy 2 sets of two DIMMs rather than those more expensive 4 DIMM sets, it’s cheaper”, to try and keep the title short. The statement/advice is true based on the idea that ‘generally’ (not always) the initial costs is lower as a pair of 2 stick sets, is normally cheaper than a 4 stick package, but in fact if they don’t play, you face return mailing fees or travel and your time, restocking fees, the fact that the store may not offer refunds, downtime from not having DRAM, etc

    Comment – “There Are Only A Few DIMM Manufacturers”

    Once again an item that was not presented as written, my line title for this, was that all too often heard “There are only a couple of companies that make DIMMS; then they all get rebranded”. My explanation may make better sense when read as an answer to my title (where I had it as ‘a couple’ rather than the editorial privilege taken by Toms to change it to say ‘a few’) . There are many that believe (mistakenly) that there are only couple of companies that manufacture DRAM and then put their name on them (rebrand).

    I appreciate the comments (and have a feeling I’ll be explaining some of these over and over, as well as others ;) )
    Reply
  • Memnarchon
    Great article, It was a good read indeed.
    Just 1 thing.
    For "13. 1.65 Volt DRAM Will Damage Your Intel CPU", I had a personal experience with previous generation DDR2 with overvloted RAM.
    Back then (with a Kentsfield CPU) as you said the MC was in Northbridge. It was fine for many years running at the rated voltage over the normal DDR2 suggestion (I think the normal was 1,8V).
    But one day after 6 years working fine, the MC from the motherboard died. The techguy that diagnosed the problem, said that this happened from years of overvolting RAM.
    My dead $200 mobo suggests that this might not be a myth afterall. :P
    But in the end this might be a problem that is left behind with DDR2 or Glenwood/Lakeport/Broadwater...
    Reply