The Notebook Review forums were hosted by TechTarget, who shut down them down on January 31, 2022. This static read-only archive was pulled by NBR forum users between January 20 and January 31, 2022, in an effort to make sure that the valuable technical information that had been posted on the forums is preserved. For current discussions, many NBR forum users moved over to NotebookTalk.net after the shutdown.
Problems? See this thread at archive.org.
 Next page →

    Dual Channel RAM Guide

    Discussion in 'Hardware Components and Aftermarket Upgrades' started by powerpack, Sep 8, 2007.

  1. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Dual Channel Ram guide.

    Why I wrote: To correct some misconceptions that are out there about what is dual channel. And to try and help with some of the questions. “Will this be Dual Channel”. This only applies to Notebooks not Desktops, they are very different, as Notebooks have two slots A & B channels and Desktops have four slots 2A & 2B channels.

    What is dual channel? The RAM bus is 64 bit, dual means using 2 (Intel) which gives you a 128 bit bus (effective), capable of twice the bandwidth. AMD is done different but the outcome is the same 64 vs 128

    Do I have it? Most current Notebook Motherboards support it, remember it is the Motherboard that determines if you can use it, not the RAM. RAM is sold in “Dual Channel kits” as both a convenience and marketing ploy.

    Do I need it? Yes/No, RAM is very fast but it does help, depending on your system, some more than others. Current Intel's take less of a performance hit not using it, AMD on the other hand makes much more use of it. Intel being 8% to 10% (est) improvement, AMD 20% to 30% (est) improvement.

    Do I need the same size RAM sticks to run in Dual Channel? Yes ( AMD at this point says yes)/No ( Intel) but you do usually want the same speed and latency as every thing clocks down to the slowest of both. Example 533Mhz CL4 mixed with 667Mhz CL5 will go with the slowest of both 533Mhz CL5 (not always with the latency) and slower than if either were matched. Same size sticks will run slightly faster symmetric RAM array (Interleaved). If the RAM is of different sizes it runs in what is called “asymmetric RAM array” ( Intel Flex Memory/ AMD does not support). The greater the disparity between your 2 sticks the greater the performance hit vs sticks of the same size. The difference seems to be approximately in proportion to the ratio of the the smaller stick x2 to the entire amount of RAM. An example with Intel, 1GB stick and 2GB stick. 2/3, 1GB(smallest stick)x2 to 3GB(the entire amount). Well 2/3 of the total improvement, of the Dual Channel running with the same size sticks (symmetric RAM array) 10%, the asymmetric RAM array example is about 6.7% so a 3.3% less than symmetric RAM array. With 512MB stick and 2GB stick, 512MB(smallest stick)x2 to 2.5GB(the entire amount). 1/2.5 4% improvement. AMD's will show greater improvements and losses, Dual/Single keep that in mind.

    What do I do? One absolute is the fastest, lowest latency is always the best unless it exceeds the the FSB speed or the supported speed for your memory, RAM will usually just clock down. You want to have an amount of RAM that minimizes how often your CPU goes to Page File whether Dual Channel or not. RAM bandwidths are likely 3000MB to 4500MB/s 60ns to 100ns latency , HDD's are 25MB to 60MB/s plus 6ms to 20ms random access times. RAM is exponentially faster than HDD. The truth is there are no absolute answers to what to do. Clearly more RAM running Single Channel or asymmetric RAM array is better than symmetric RAM array with out enough to avoid the Page File.

    Are some brands better than other? Yes/NO, all RAM of a rated speed and latency performs the same regardless of brand. Some of the expensive brand name RAM offers things like heat spreaders. Reducing heat is always a good thing but not always necessary and does not affect speed. Some people who overclock like these features. Buy from a reputable retailer and get a lifetime warranty and you will be fine with any RAM.

    Conclusions: With Intel's I don't think people should worry as much about their RAM configurations. Only up to 10% difference. Make sure you have enough, AMD people you should care a little more up to 30% difference but not at the expense of too little RAM. Those with IGP's might want to consider symmetric RAM array more than others.

    Let me try and give an example to try and illustrate the difference between the Single Channel and the two Dual Channels listed above. It is simplified but I think valid. We have two bottles exactly the same.

    For Single Channel we fill both with 1 liter We turn one upside down to drain as soon as it is empty with no delay we turn the second one upside down until empty. We have timed it and we know our quantity. So we can get a ounce/s number

    Symmetric RAM array some call “true” Dual Channel. Same two bottles fill 1 liter each, turn them upside down at the same time until empty. We have timed, to get the ounces/s, I am going to guess the ounce/s is going to be double the previous (don't laugh I know Einstein is rolling in his grave).

    Finally the big controversy, Asymmetric RAM array. We fill one with 1 liter and the other with 2 liters turn them upside down until empty. We have timed, to get ounces/s.

    What does that tell us? Well if you have 128 bit bus with 3GB's it is not Single Channel. It is not the same as symmetric bottle array, it is less ounces/s but it is more than Single Channel because it uses the 128 bit.

    Here are some links that support some of what I say, nothing spells it out, which is why I wrote and got results on my own. These links actually disagree on some terminology. They are not notebooks, I could not find anything on notebooks but some info is applicable.

    Link #1: Intel desktops. RAM Channel explanations.
    Link #2: Describes RAM Channels on a server.
    Link #3: Discusses Intel's "Flex Memory"


    I would like to thank John Ratsey as he provided me with the Intel benchmarks I use, I have a very small sample but it is the best I could do.
    If anyone has a Turion and can run matching sticks and different size sticks please PM me if you would like to help me.
     
  2. allan_huang

    allan_huang Notebook Deity

    Reputations:
    25
    Messages:
    1,030
    Likes Received:
    0
    Trophy Points:
    55
    2 BIG ups for you.
    Should definitely be stickied.
    There are quite a few people out there asking about ram all the time.
    +1 for you!
     
  3. Thomas

    Thomas McLovin

    Reputations:
    1,988
    Messages:
    5,253
    Likes Received:
    0
    Trophy Points:
    205
    perhaps expanding it would be good and all but its good!
    +Rep
     
  4. Jalf

    Jalf Comrade Santa

    Reputations:
    2,883
    Messages:
    3,468
    Likes Received:
    0
    Trophy Points:
    105
    A lot of half explanations here... Mind if I expand a bit on some of them? ;)

    Using 2 what?
    Dual channel means that the memory controller has two "connections" to the RAM (typically done on desktops by having each connection handle two RAM slots each (out of the four that are available on the motherboard).
    What has AMD/Intel done differently? They don't (always) make the motherboards. If you don't know what they've made differently, and it doesn't make a difference, don't bother mentioning it. :)

    And what it actually means is that you just get two identical RAM sticks, which is guaranteed to work because dual-channel relies on both channels having similar memory configurations.

    Hmm, I didn't know about the asymmetric stuff. Interesting. Keep in mind though, that not all motherboards support that. (I've seen it work on laptops. I know a year or two, no desktops could do it. Don't know if it's 100% standard now, or if it's primarily a laptop thing, to compensate for only having two ram slots)

    For asymmetric mode specifically? If you're still talking about the "AMD is more reliant on dual-channel in general thing, then remove it from this section, since you already mentioned it, and it's the same effect regardless of symmetric/asymmetric modes. (AMD just needs more memory bandwidth)
    If you're talking about a specific effect of using asymmetric with AMD, you should explain that. :)

    And RAM has ~100ns access time. You should mention that too. :)

    Or to put it another way, we have one 64-bit connection going out to the 1GB stick, and another 64-bit connection going out to the 2GB stick.
    So depending on what part of memory is being read, we'll get an effective connection of either 64 or 128 bit. If they were symmetric, the data would be distributed evenly, so you'd always get to use the 128-bit interface, but now, more data can be stored in one side than the other. So obviously, we'll sometimes have to refer to that side only. And every time we do, we drop down to single-channel speeds.

    Slow down a bit, and proof-read ;)
    Makes it a lot easier to follow your argument if you fix the grammatical errors and half sentences. :)

    Also, some (links to) tests or benchmarks would be nice. Or graphs or diagrams.

    Anyway, nice work :)
     
  5. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Jalf: you little elf! thanks for the input, what you said, I will think about and consider.
    This is in fact where much of the confusion comes from. Desktops! They require 2 of their 4 to match. Well notebooks only have 2 so consider each slot already matched, therefor you don't have to match. that is both a fact and a common misconception, which is what I am trying to dispel.

    Thanks for the input.

    Jalf: If you want to edit that is fine with me, as that this is only about getting accurate info out there.
     
  6. Jalf

    Jalf Comrade Santa

    Reputations:
    2,883
    Messages:
    3,468
    Likes Received:
    0
    Trophy Points:
    105
    A fact *and* a misconception? o_O
    What does that mean? :p
    Obviously, if their slots are already matched, you should be able to run dual-channel with one single ram stick. Is that the case?
    Do you have a link to other sources explaining how this works?

    No thanks. :)
    I just wanted to point out what you could improve on your article. :)

    It's a good read though. Nice job.
     
  7. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Well at least right now if you Google "asymmetric RAM array dual channel" 1st one you will get is this post. Kind of makes it harder for me to research this topic and correct or expand. Wish I checked my facts better to start.
     
  8. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Bumped! Please comment! Even if you disagree. So often it comes up on this board!
     
  9. BenArcher

    BenArcher Notebook Consultant

    Reputations:
    20
    Messages:
    189
    Likes Received:
    0
    Trophy Points:
    30
    Should be made a sticky.

    Hopefully it could stop most of the questions happening again & again & again.

    PS the ram access time varies depending on what operation is being performed thats why there are multiple numbers in the timings. In general the total latency for an intel system will be between 60ns & 100ns (total latency being once something is requested how long it takes to get the info). AMD systems have lower latency becasue on board memory controllers.
     
  10. prashanthm

    prashanthm Notebook Consultant

    Reputations:
    20
    Messages:
    136
    Likes Received:
    0
    Trophy Points:
    30
    That's a nice post powerpack..., repped
     
  11. John Ratsey

    John Ratsey Moderately inquisitive Super Moderator

    Reputations:
    7,197
    Messages:
    28,841
    Likes Received:
    2,165
    Trophy Points:
    581
    It's now promoted to be a minor thread in the sticky, which makes it easier to find.

    John
     
  12. Jitsun

    Jitsun Notebook Guru

    Reputations:
    0
    Messages:
    51
    Likes Received:
    0
    Trophy Points:
    15
    Reading about this, I'm wondering whether 2GB would be a wiser idea over my initial thought of upgrading to 3GB.

    Although I would like the extra ram for other work, I don't want it interfering with casual gaming.

    Would the difference between single channel (1 GB & 2 GB configuration) and dual (2GB) be a problem?.

    .. Although the cost is a little more then I wanted to spend at this point, should I just upgrade to 4GB?.

    Laptop would be running Windows XP Pro. I know it'll only recognize 3.5GB but wonder whether that would make a difference. (HP 8510p)

    [P.S. Yes I've read the first post. I want someone's personal advice.]
     
  13. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    1GB + 2GB is Dual Channel because it makes use of 128 bit bus. Intel calls flex mode but they do call it Dual channel. I would not hesitate to use this configuration as stated in post 3% less than 2X1GB, also I have found more info from Intel and the first 2GB will be addressed 1st so I suspect that once you start running a game the information will find it's way towards the 1st 2GB which runs at "true" Dual Channel while the last 1GB is in Single Channel.

    But I will mention running XP I highly doubt you need more than 2GB's. But I don't don't really see a downside with 3%.
     
  14. Jalf

    Jalf Comrade Santa

    Reputations:
    2,883
    Messages:
    3,468
    Likes Received:
    0
    Trophy Points:
    105
  15. John Ratsey

    John Ratsey Moderately inquisitive Super Moderator

    Reputations:
    7,197
    Messages:
    28,841
    Likes Received:
    2,165
    Trophy Points:
    581
    Please see this thread for some bandwidth measurement results for the Intel 965M chipset and different RAM configurations. Cast the theory aside and let the measurements speak for themselves.

    I can't speak for the AMD CPU RAM performance but I have some results for the ATI 1250M chipset (and Intel T2250 CPU - see Samsung R20 review for full hardware details). In summary:

    RAM Sandra Memory Bandwidth
    1 x 512MB PC4200 ... 2657 / 2617
    2 x 512MB PC4200 ... 2960 / /2975
    2 x 512MB PC5300 ... 3013 / 3015
    1 x 1GB PC4200 ... 2670 / 2687
    1 x 1GB PC5300 ... 2871 / 2860
    1.5Gb PC4200 ... 3019 / 3013
    2 x 1GB PC4200 ... 2930 / 2935
    1 x 2GB PC5300 ... 2922 / 2904
    2.5GB PC4200 ... 2980 / 3007
    2.5GB PC5300 ... 3090 / 3110

    What we see is that any combination of two modules gives a bandwidth within a close range. A low capacity single module tends to be a bit slower. It is possible that another factor in the equation is the number of chips on the module and whether it is single or double sided.

    John
     
  16. MrFugi

    MrFugi Notebook Enthusiast

    Reputations:
    0
    Messages:
    17
    Likes Received:
    0
    Trophy Points:
    5
    Rep'd! Thanks a lot for all the information, it definitely cleared things up for me because I'm looking to upgrade my Inspiron 9400 memory, although I still can't find a good deal in Canada for a 2GB Single Stick :(

    Anyone have a suggestion? (Sorry, I don't want to detract from the purpose of the thread)
     
  17. John Ratsey

    John Ratsey Moderately inquisitive Super Moderator

    Reputations:
    7,197
    Messages:
    28,841
    Likes Received:
    2,165
    Trophy Points:
    581
    Did you look through the Hot RAM Deals thread?

    John
     
  18. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    This is what it actually says.

    Intel® Desktop Boards based on the following chipsets support Flex Mode:

    Q35
    P35
    G33
    P965
    G965
    Q965
    Q963

    Your use of the word “only” is deceptive, as this is only talking about Desktops not Notebooks so it does not exclude them, it does not mention them.

    This is from one of John's tests, it is the 1GB + 2GB test.

    Logical/Chipset 1 Memory Banks
    Bank 0 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 1 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 4 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 5 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Shared Memory : 8MB
    Channels : 2
    Memory Bus Speed : 4x 333MHz (1332MHz data rate)
    Width : 64-bit

    See where it says Channels? 2 (Dual), 2X64=128

    You did do some digging, I know, I found that document also (1st link at the bottom of guide for a couple of days now). There is very little info on this subject and nothing on notebooks.
    As far as AMD goes you are correct that I do not know if it works. But at this point I have found nothing to indicate either way. And am considering my options to obtain a 512MB stick and check.

    Some updates on my reading since I wrote the guide. The document Jalf (and I) have linked from Intel says in flex memory (3GB) that the lowest (front) part of the RAM is the matched and the higher (back) part is the solo. So in the real world I believe 3GB likely will function better than 2GB. If you load an application or game it will fill the lower part of RAM 1st (or at the very least will migrate there). So you will be running in dual channel. If it had to go to the 3rd GB the 2GB would have to go to page file. The 3rd GB in single channel is much faster than the 3rd GB page file. Not certain on this yet just a thought in progress.
     
  19. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    Also, we need to remember that the newer Crestline chipset and the older Calistoga handle asynchronous dual channel differently. It seems that this "Flex Mode" is the newer handling of Asynchronous Dual Channel, and from what I've seen it provides no real disadvantages over normal Dual Channel. However the older Calistoga chipset does not support "Flex Mode", and handles asynchronous dual channel as described (it will dual channel the first 1GB stick and half of the second 2GB stick, and the rest of the 2GB stick will run in single channel).
     
  20. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Yes! Any input you have I would like to know as there is noinfo out there! I think maybe you misunderstood Jalf's post! What are you saying?
     
  21. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    I'm saying that "Flex Mode" is different from asynchronous dual channel. Any Intel chipset that supports dual channel also supports asynchronous dual channel (which I've explained before and is similar to RAID 0 between two disks of different capacities.) "Flex Mode" is supported by the newer Crestline chipset (and the newer desktop chipsets, but I don't know as much about them), and is more intellegent than basic asynchronous dual channel. A full description is on the Intel page linked above, but to the best of my knowledge is has no disadvantages over regular dual channel.

    I don't know anything about the AMD side though, sorry.
     
  22. Jalf

    Jalf Comrade Santa

    Reputations:
    2,883
    Messages:
    3,468
    Likes Received:
    0
    Trophy Points:
    105
    odin243: Got a few links on this stuff?
    The one I linked to, that's the "old-fashioned" way then? (Then why do they call it flex mode, if flex mode only refers to the newest chipsets?)

    And if flex mode is different from asynchronous dual-channel, someone must have a link explaining async dual-channel. You must have read about it somewhere... (Unless someone can dig up some information saying these things are different, I'm inclined to assume that flex mode == asynchronous dual channel)
     
  23. Les

    Les Not associated with NotebookReview in any way

    Reputations:
    4,706
    Messages:
    5,391
    Likes Received:
    1
    Trophy Points:
    0
    Can we add to this?

    I run 64 Bit Vista with 4Gb OCZ RAM. My sidebar CPU monitor shows that, on average, I am only ever using 30% of this.

    If I eliminate pagefile.sys, will this enhance the performance at all? Can I eliminate it totally?

    Also, is there a way to ensure basic program run totally inside of the ram (since I have so much) to make them work and respond much faster.

    Yes, being new to the 4Gb of RAM thing, I am just looking for RAM specific tweaks that will allow me to get the most out of the RAM I have.

    Def Rep2U here!!
     
  24. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    When I want to play with page file I do it in System Properties. I have played with adjusting but have gone back to using. I don't see much of a downside to having it, but I don't really know so any comment is cool with me.
     
  25. Jalf

    Jalf Comrade Santa

    Reputations:
    2,883
    Messages:
    3,468
    Likes Received:
    0
    Trophy Points:
    105
    flamenko: That should probably go in a separate thread (it has nothing to do with single- vs dualchannel RAM)

    I once wrote a long post explaining virtual memory and the pagefile, you might try to dig it up. I don't really like how it turned out though, so might rewrite it some time.
    Anyway, the short version is this:
    Disabling the pagefile doesn't really achieve anything. Think of the pagefile as a safety net. It's there for data to spill over into when you run out of RAM. If you never run out of RAM, the pagefile (almost) doesn't get used, so it doesn't matter that it's there.
    As for ensuring programs run in RAM? They do all by themselves. The CPU can only work with data in RAM. If something is in the pagefile, it has to pause, pull the data from pagefile to RAM, and then resume. So simply by running a program, all the data it touches automatically gets pulled into RAM.
     
  26. Les

    Les Not associated with NotebookReview in any way

    Reputations:
    4,706
    Messages:
    5,391
    Likes Received:
    1
    Trophy Points:
    0
    Tx Jalf..Repped

    You have actually answered my questions, quick and simple.
     
  27. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    No, the one you linked to is indeed the newer way, that's why it's only present on newer chipsets (only the 965PM and 965GM support it, to the best of my knowledge). It's been a while since I looked into this, but if I remember correctly asynchronous dual channel isn't really a "feature", it's just what happens if you try to run two different sticks in a dual channel mobo (prior to flex mode). I'm not sure it's ever been well documented how this works on mobile mobos, though I can find some older examples of it on desktop mobos. For example, here we can see the huge bandwith disadvantage to using asynchronous dual channel over regular dual channel:
    http://www.digit-life.com/articles2/digests/dram2k4.html
    Those are tests with early DDR2 RAM I believe.

    I will try to dig up some benchmarks to show the practical difference between Flex Mode and the older Asynchronous Dual Channel. As to where I first read about it, I'm trying to find it, but it was in one of the teasers for Santa Rosa a while back, when Intel said they were releasing a new type of dual channel for mobile chipsets that would allow you to use different size SODIMMS.
     
  28. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    The long and the short of the article you linked, is the asynchronous it refers to is between the FSB and RAM speed not the size of the RAM sticks so a completely different issue. All it is trying to demonstrate is that RAM that is faster than your FSB returns minimal bandwidth improvement and higher latency vs matched FSB and RAM. At the time this was written the asynchronous refered to a slower FSB, where as now for example with Santa Rosa platform FSB is 800Mhz and RAM is 667Mhz which is also asynchronous but reversed. Is there penalty for not being synchronous in this way? I don't know I suspect maybe slight because clock cycles are not in synch but I do not know. Trying to figure what is and is not Dual Channel keeps me busy enough right now.
     
  29. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    Sorry, you're right, that article is addressing a different issue. Doing some more digging, here are some benchmarks related to the early development of dual channel. What is called "Flexible Dual Channel" in the article is SIS' solution to asynchronous dual channel, and it's the same solution employed by pre 945 intel chipsets I believe.
    http://www.sis.com/elibrary/elibrary_index03_000006.htm


    Edit: Okay, after further digging, I've come up with this. If you read through memory configuration guides from 915 and 945 Intel chipsets, you'll see descriptions like the following:
    However, the newer descriptions for things like the 965 boards are phrased like this:
    I'm still looking for hard benchmarks of the Flex Memory technology, but I think that the above explanations highlight the theoretical difference well.
     
  30. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Here is a Link from my OP. It calls it Single Channel (Asymmetric), Engineers and technical writers are not the standard barriers of language thank God. It appears definitions have changed, if that is your point than I would say you are correct. If I read your Dual Channel Asymmetric correct it refers two using both channels on the MoBo that is where it gets Dual, It explains how it addresses it and that is clearly “Single Channel” until “Flex Mode” was developed there was no reason to describe any different. But with the advent of the new technology the naming conventions and what they meant needed to change. Look at my link. If you disagree with how I identify the technologies so be it, I have thought about it and at this point am not looking at any one document as your own quotes show Intel even changed. If in time things are standardized and become uniform and in conflict with what I wrote I will adjust.

    John Ratsy did post benchmarks earlier in the thread.
     
  31. sivakrishnamannem

    sivakrishnamannem Notebook Enthusiast

    Reputations:
    0
    Messages:
    35
    Likes Received:
    0
    Trophy Points:
    15
    hey this is a interesting thread, i want to upgrade my ram from 1 gb to 2b, i have a one 1 gb stick which is ddr 2 pc2 4300 533 mhz, do i need to get a another same one or can i club a 667 mhz with this, please advice me...
     
  32. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    Whichever you get, it will downclock to 533MHz, so just get whichever is cheaper.
     
  33. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Only concern is if the higher latency of 667Mhz CL5 makes both run at CL5 if so get 533Mhz CL4.
     
  34. Jalf

    Jalf Comrade Santa

    Reputations:
    2,883
    Messages:
    3,468
    Likes Received:
    0
    Trophy Points:
    105
    Ah, got it. Thanks odin243, for digging up those articles.

    Technically speaking, single-channel is when only one channel is populated with a RAM block.
    If both channels are populated, but with different amounts of RAM, we enter asynchronous mode, where channel A covers all addresses from 0 up to (size of A-1), and B covers addresses from (size of A) to (size of A + B -1)
    That will almost always give you single-channel bandwidth in practical usage, although technically speaking, both channels are active. (and benchmark tools may report two channels)
    This is called either dual- or single-channel asynchronous mode depending on who you ask, and performs no better than "strict" single-channel mode where only one channel is populated.

    Flex mode (which seems to be implemented on newer Intel desktop boards, not at all on AMD boards, and unknown on Intel mobile boards) interleaves the addresses (so conceptually, all even addreses are located on channel A, and uneven on channel B, up until one channel runs out of memory, after which it is back to single-channel mode.
    So for all memory accesses below a certain address (determined by twice the amount of memory in the "smallest" channel), you get dual-channel performance, and above that, single-channel.

    And finally, full dual-channel simply doesn't need the single-channel fallback that flex mode uses, because both channels contain the same amount of RAM, so no matter the memory address, you'll run in dual-channel.

    Cool then.
     
  35. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    Yup, that's a pretty good explanation, only one thing I would say:
    While this may be true for mobile mobos, it's not true in general. With older DDR2 Dual Channel (not the original ones, those wouldn't even post with an uneven RAM configuration), uneven RAM configurations would run in only a single channel. I don't think that was ever applied to mobile mobos, but I'm not sure.


    Oh no, I don't really disagree with anything in your post, I'm sorry if it seemed that way. I was just trying to see if I could clarify how exactly asymmetric dual channel works on different chipsets. As for benchmarks, I'm still looking. John's were on a 1250m chipset I believe. I'm trying to find some that illustrate the difference in performance between the 915M-945M chipsets (which used the same general type of memory controller I believe), and the newer 965M chipset, which should have a big advantage in the area of asymmetric dual channel.
     
  36. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Jalf, very very well written but consider these points please.

    No look at this Link and scroll down several lines, you will see that statement is incorrect. If Intel's Document matters.

    Asymmetric is the term used on odins documents and the more recent Intel document calls it Single Channel (Asymmetric). And second quote describes what Intel call....
    Clearly contradicts what you stated. And is more recent than Intel document that contradicts it.

    Unknown on Intel Mobile Boards? John Ratsy has the benchmarks, this is one you have not conceded yet. Here is a quote from odin's link
    This is where “flex mode” comes from 2X64. This is how SiSoft reports 965 chipset so it is “flex” my AMD reports 1X128 and does not support so to assume a benchmark miss reads as you imply is unlikely given this info.
     
  37. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    To be fair powerpack, we've seen no benchmarks showing if Intel Flex Memory has benefits on mobile chipsets, nor if it's even present on mobile chipsets.
     
  38. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Dual, Flex, Single
    Here is some detail to support what I say that your SiS document talks about.
    here we have 1X2GB +1X1GB look 2 channels 64 bit exactly what the document talks about for "flex mode"

    Logical/Chipset 1 Memory Banks
    Bank 0 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 1 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 4 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 5 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Shared Memory : 8MB
    Channels : 2
    Memory Bus Speed : 4x 333MHz (1332MHz data rate)
    Width : 64-bit
    Performance Acceleration Technology : No
    Memory Controller in Processor : No
    Maximum Memory Bus Bandwidth : 21312MB/s (estimated)

    Separate memory controllers, his has been posted earlier in this thred.

    Here is the AMD 2X1GB which does not support and looks as that Sis link explained.

    Logical/Chipset 2 Memory Banks
    Bank 0 : 1GB DDR2-SDRAM 4.0-4-4-11 (tCL-tRCD-tRP-tRAS) CR2
    Bank 1 : 1GB DDR2-SDRAM 4.0-4-4-11 (tCL-tRCD-tRP-tRAS) CR2
    Channels : 1
    Bank Interleave : 2-way
    Memory Bus Speed : 2x 302MHz (604MHz data rate)
    Width : 128-bit
    Memory Controller in Processor : Yes
    Cores per Memory Controller : 2 Unit(s)
    Maximum Memory Bus Bandwidth : 9664MB/s (estimated)

    One channel 128

    What more would you like too see? If you would like to see the entire benchmarks I will post. the numbers I have are from these as stated in OP.
     
  39. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    <STRIKE>John's post that you've quoted is enlightening, but as I said before that's on a 1250m chipset,</STRIKE> and the other benches you've listed appear to be on a desktop board and an AMD board. I was saying that specifically on the 965M chipsets, we've yet to see real world confirmation of Intel Flex Memory or it's benefits.

    Edit: Yes, I agree with all those benchmarks, but I'm still looking for some performed on a 965M chipset. If you know of any, could you please post/link them?
     
  40. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    odin the AMD is mine and it is mobile, John has told me his is a Zepto notebook with a T7300 so it is not a desktop board. I think Johns are what you ask, if I am mistaken let me know, look at John's computers, notebooks. Jalf's doubts and your questions are exactly why I wote this. A total lack of info on notebooks, different words that disagree, general confusion. John had a Intel notebook and different size sticks. He ran tests and they supported what you question. I started out thinking that Asymmetric RAM was crap but changed midway so I understand doubt. Those are mobile boards.

    Edit: If John is running Santa Rosa which he is then he is running the chipset you ask for? No desktop Santa Rosa I know about?
     
  41. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    Hmm, I was assuming those tests weren't on a 965M board, based on the following (from his post):
     
  42. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    At the beginning of that post he links in about the first three words to tests for 965m as you asked. I think we are correct as I have always said, "if I'm wrong, let me have it" give it a look. Heck I have AMD if you Intel guys don't have flex HaHa! But I think you do.

    Edit: odin if anything I say in the original post is wrong let me know as I can change it. I know I am working with half facts and poor documentation. I would like it to be helpful and correct, so if you find things or think it is wrong let me know. But so far I can't concede anything of what I wrote yet. Except I have conceded things to Jalf but I am not going to mention those.
     
  43. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    I know he links to it, but if you look at the benches in the thread he links to, you'll see they don't appear in the list later on in his post. The linked benches are nice, and for reference they say:
    1 x 1GB: 3685/3671
    2 x 1GB: 3532/3559
    1 x 2GB: 4012/4021
    3GB: 3919/3879

    However those benches appear suspect, because there is no way that a single 2GB stick has 10% better bandwith than two 1GB sticks. I'm guessing they're labeled incorrectly, but even if not, I'd still like to see another set of benches on the 965M chipset (you really need at least two reliable sets to draw any conclusions).
     
  44. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    You asked for it! And I do not think your request for a second set of benchmarks are valid as what I have to work with is so small. but John did run these many times and all were in expected standards. So here they are!

    1st single, 2nd Asymmetric Dual Channel (new standard), 3rd Dual Channel!

    SiSoftware Sandra

    Benchmark Results
    Int Buff'd iSSE2 Memory Bandwidth : 3658 MB/s
    Float Buff'd iSSE2 Memory Bandwidth : 3674 MB/s
    Results Interpretation : Higher index values are better.

    Int Buff'd iSSE2 Memory Bandwidth
    Assignment : 3626 MB/s
    Scaling : 3616 MB/s
    Addition : 3671 MB/s
    Triad : 3722 MB/s
    Data Item Size : 16 byte(s)
    Buffering Used : Yes
    Offset Displacement Used : Yes
    Bandwidth Efficiency : 57% (estimated)

    Float Buff'd iSSE2 Memory Bandwidth
    Assignment : 3681 MB/s
    Scaling : 3679 MB/s
    Addition : 3698 MB/s
    Triad : 3639 MB/s
    Data Item Size : 16 byte(s)
    Buffering Used : Yes
    Offset Displacement Used : Yes
    Bandwidth Efficiency : 57% (estimated)

    Performance Test Status
    Run ID : ZEPTO-6024W on 22 August 2007 at 19:11:03
    Platform Compliance : Win32 x86
    Memory Used by Test : 512MB
    NUMA Support : No
    SMP (Multi-Processor) Benchmark : No
    Total Test Threads : 1
    Multi-Core Test : Yes
    SMT (Multi-Threaded) Benchmark : No
    Processor Affinity : P0C0T0
    System Timer : 14MHz
    Page Size : 4kB
    Use Large Memory Pages : No

    Features
    SSE Technology : Yes
    SSE2 Technology : Yes
    SSE3 Technology : Yes
    Supplemental SSE3 Technology : Yes
    SSE4 Technology : No
    EMMX - Extended MMX Technology : No
    SSE4A Technology : No
    HTT - Hyper-Threading Technology : No

    Chipset 1
    Model : Inventec Corp Mobile PM965/GM965/GL960 Express Processor to DRAM Controller
    Revision : A4
    Front Side Bus Speed : 4x 200MHz (800MHz data rate)
    Width : 64-bit
    Maximum Bus Bandwidth : 6400MB/s (estimated)

    Logical/Chipset 1 Memory Banks
    Bank 0 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 1 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Shared Memory : 8MB
    Channels : 1
    Memory Bus Speed : 4x 333MHz (1332MHz data rate)
    Width : 64-bit
    Performance Acceleration Technology : No
    Memory Controller in Processor : No
    Maximum Memory Bus Bandwidth : 10656MB/s (estimated)

    Performance Tips
    Notice 5405 : System bandwidth appears FSB limited. Attempt to increase FSB.
    Notice 5008 : To change benchmarks, click Options.
    Notice 5004 : Synthetic benchmark. May not tally with 'real-life' performance.
    Notice 5006 : Only compare the results with ones obtained using the same version!
    Warning 5010 : Cannot use Large Memory Pages due to lack of privileges.
    Warning 2544 : System/Video shared memory greatly reduces performance. Use external video card.
    Tip 2 : Double-click tip or press Enter while a tip is selected for more information about the tip.


    SiSoftware Sandra

    Benchmark Results
    Int Buff'd iSSE2 Memory Bandwidth : 3847 MB/s
    Float Buff'd iSSE2 Memory Bandwidth : 3859 MB/s
    Results Interpretation : Higher index values are better.

    Int Buff'd iSSE2 Memory Bandwidth
    Assignment : 3773 MB/s
    Scaling : 3752 MB/s
    Addition : 3926 MB/s
    Triad : 3937 MB/s
    Data Item Size : 16 byte(s)
    Buffering Used : Yes
    Offset Displacement Used : Yes
    Bandwidth Efficiency : 60% (estimated)

    Float Buff'd iSSE2 Memory Bandwidth
    Assignment : 3823 MB/s
    Scaling : 3832 MB/s
    Addition : 3922 MB/s
    Triad : 3862 MB/s
    Data Item Size : 16 byte(s)
    Buffering Used : Yes
    Offset Displacement Used : Yes
    Bandwidth Efficiency : 60% (estimated)

    Performance Test Status
    Run ID : ZEPTO-6024W on 22 August 2007 at 21:34:45
    Platform Compliance : Win32 x86
    Memory Used by Test : 512MB
    NUMA Support : No
    SMP (Multi-Processor) Benchmark : No
    Total Test Threads : 1
    Multi-Core Test : Yes
    SMT (Multi-Threaded) Benchmark : No
    Processor Affinity : P0C0T0
    System Timer : 14MHz
    Page Size : 4kB
    Use Large Memory Pages : No

    Features
    SSE Technology : Yes
    SSE2 Technology : Yes
    SSE3 Technology : Yes
    Supplemental SSE3 Technology : Yes
    SSE4 Technology : No
    EMMX - Extended MMX Technology : No
    SSE4A Technology : No
    HTT - Hyper-Threading Technology : No

    Chipset 1
    Model : Inventec Corp Mobile PM965/GM965/GL960 Express Processor to DRAM Controller
    Revision : A4
    Front Side Bus Speed : 4x 200MHz (800MHz data rate)
    Width : 64-bit
    Maximum Bus Bandwidth : 6400MB/s (estimated)

    Logical/Chipset 1 Memory Banks
    Bank 0 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 1 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 4 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 5 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Shared Memory : 8MB
    Channels : 2
    Memory Bus Speed : 4x 333MHz (1332MHz data rate)
    Width : 64-bit
    Performance Acceleration Technology : No
    Memory Controller in Processor : No
    Maximum Memory Bus Bandwidth : 21312MB/s (estimated)

    Performance Tips
    Notice 5405 : System bandwidth appears FSB limited. Attempt to increase FSB.
    Notice 5008 : To change benchmarks, click Options.
    Notice 5004 : Synthetic benchmark. May not tally with 'real-life' performance.
    Notice 5006 : Only compare the results with ones obtained using the same version!
    Warning 5010 : Cannot use Large Memory Pages due to lack of privileges.
    Warning 2544 : System/Video shared memory greatly reduces performance. Use external video card.
    Tip 2 : Double-click tip or press Enter while a tip is selected for more information about the tip.


    SiSoftware Sandra

    Benchmark Results
    Int Buff'd iSSE2 Memory Bandwidth : 3980 MB/s
    Float Buff'd iSSE2 Memory Bandwidth : 4015 MB/s
    Results Interpretation : Higher index values are better.

    Int Buff'd iSSE2 Memory Bandwidth
    Assignment : 3951 MB/s
    Scaling : 3914 MB/s
    Addition : 4030 MB/s
    Triad : 4027 MB/s
    Data Item Size : 16 byte(s)
    Buffering Used : Yes
    Offset Displacement Used : Yes
    Bandwidth Efficiency : 62% (estimated)

    Float Buff'd iSSE2 Memory Bandwidth
    Assignment : 3941 MB/s
    Scaling : 3959 MB/s
    Addition : 4102 MB/s
    Triad : 4058 MB/s
    Data Item Size : 16 byte(s)
    Buffering Used : Yes
    Offset Displacement Used : Yes
    Bandwidth Efficiency : 63% (estimated)

    Performance Test Status
    Run ID : ZEPTO-6024W on 22 August 2007 at 19:33:08
    Platform Compliance : Win32 x86
    Memory Used by Test : 512MB
    NUMA Support : No
    SMP (Multi-Processor) Benchmark : No
    Total Test Threads : 1
    Multi-Core Test : Yes
    SMT (Multi-Threaded) Benchmark : No
    Processor Affinity : P0C0T0
    System Timer : 14MHz
    Page Size : 4kB
    Use Large Memory Pages : No

    Features
    SSE Technology : Yes
    SSE2 Technology : Yes
    SSE3 Technology : Yes
    Supplemental SSE3 Technology : Yes
    SSE4 Technology : No
    EMMX - Extended MMX Technology : No
    SSE4A Technology : No
    HTT - Hyper-Threading Technology : No

    Chipset 1
    Model : Inventec Corp Mobile PM965/GM965/GL960 Express Processor to DRAM Controller
    Revision : A4
    Front Side Bus Speed : 4x 200MHz (800MHz data rate)
    Width : 64-bit
    Maximum Bus Bandwidth : 6400MB/s (estimated)

    Logical/Chipset 1 Memory Banks
    Bank 0 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 1 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 4 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Bank 5 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
    Shared Memory : 8MB
    Channels : 2
    Memory Bus Speed : 4x 333MHz (1332MHz data rate)
    Width : 64-bit
    Performance Acceleration Technology : No
    Memory Controller in Processor : No
    Maximum Memory Bus Bandwidth : 21312MB/s (estimated)

    Performance Tips
    Notice 5405 : System bandwidth appears FSB limited. Attempt to increase FSB.
    Notice 5008 : To change benchmarks, click Options.
    Notice 5004 : Synthetic benchmark. May not tally with 'real-life' performance.
    Notice 5006 : Only compare the results with ones obtained using the same version!
    Warning 5010 : Cannot use Large Memory Pages due to lack of privileges.
    Warning 2544 : System/Video shared memory greatly reduces performance. Use external video card.
    Tip 2 : Double-click tip or press Enter while a tip is selected for more information about the tip.


    There it is! 965m!
     
  45. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    Thank you. Those benchmarks confirm what I've suspected about bandwith in with the Flex Memory tech.
     
  46. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Good/Bad I'm right/wrong let me know. ;)
     
  47. odin243

    odin243 Notebook Prophet

    Reputations:
    862
    Messages:
    6,223
    Likes Received:
    0
    Trophy Points:
    205
    It confirms that there is no practical downside to an asymmetrical RAM array on the 965 chipset. The difference in bandwith is pretty much negligable, and the effect of memory bandwith on application performance is small enough to make it not even worth mentioning. Since you'll pretty much always have at least 1GB of dual-channel space, and since that's enough for your active apps (normally), the average person no longer needs to worry about symmetrical dual channel on Intel boards.
     
  48. Lithus

    Lithus NBR Janitor

    Reputations:
    5,504
    Messages:
    9,788
    Likes Received:
    0
    Trophy Points:
    205
    While you guys are arguing over this, Bob over there is still wondering where the floppy goes on these new-fangled things.
     
  49. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Lithus: Thanks for all the support when I needed you!
     
  50. powerpack

    powerpack Notebook Prophet

    Reputations:
    7,101
    Messages:
    5,757
    Likes Received:
    0
    Trophy Points:
    0
    Why did you pull my teeth? As I said in prior posts, it seems to me if you have a 3GB configuration it will act like a 2GB configuration and hit the pagefile less, but I do not know! I think most apps front load in RAM as all think they are the most important! Just my thoughts.
     
 Next page →