Core i7 Multipliers - Q&A

edited December 2008 in Hardware
I was more interested in the avaiable Multipliers for Overclocking (in the BIOS).

But thanks for posting that information.

Comments

  • KhaosKhaos New Hampshire
    edited December 2008
    I was more interested in the avaiable Multipliers for Overclocking (in the BIOS).

    But thanks for posting that information.
    Sorry, I did post that information but I may have muddled it in too much geek speak.

    The Core i7 920 has a stock multiplier of 20x and you can adjust it DOWN, so your range in the BIOS will be like 14x to 20x depending on the motherboard.

    The Core i7 940 has a stock multiplier of 22x and you can adjust it down as well, so same concept but the max multiplier will be 22x instead of 20x.

    Edit: More test results are coming out that seem to contradict some of the previous statements made by people about memory limitation, so a portion of this post was invalidated. My i7 needs to get here so I can stop relying on others. >.<
  • edited December 2008
    Misexplained myself once more, curious about avaiable multipliers for QPI, Uncore and Memory for the 3 CPU's. I'm trying to make my purchase decision based on that.
  • KhaosKhaos New Hampshire
    edited December 2008
    Misexplained myself once more, curious about avaiable multipliers for QPI, Uncore and Memory for the 3 CPU's. I'm trying to make my purchase decision based on that.
    Ohhhh, I get you.

    The only locked multiplier that you need to consider when choosing a processor is the CPU-Core; I feel fairly safe in making that statement.
    Khaos wrote:
    The QPI multiplier is a function of the motherboard, and is thus adjustable upward or downward for all models. The BIOS simply sets a default based on the processor model. Of course, 920 and 940 models are not guaranteed to function at a higher QPI -- although it seems that most do so just fine. The binning process seems to be most focused on final CPU and Uncore frequencies attainable.
    Khaos wrote:
    The interesting thing to note here is that adjusting the uncore multiplier on i7 processors does not yield any performance gain, and Intel's own default settings seem to disregard this rule of uncore = 2*mem.
  • edited December 2008
    My actual doubts are related to being able to use DDR3 2000Mhz on a 920 without "hurting" the OC on the CPU, not sure which Memory and Uncore multipliers avaiable in the 920.
  • KhaosKhaos New Hampshire
    edited December 2008
    My actual doubts are related to being able to use DDR3 2000Mhz on a 920 without "hurting" the OC on the CPU, not sure which Memory and Uncore multipliers avaiable in the 920.
    Your memory choice won't hurt your CPU overclock. With memory multipliers all the way down to 6x, you will be fine with any DDR3 that is capable of 1200Mhz or so.

    What you do need to look at with you memory choice is the rated voltage:

    The Core i7 processors, due to their new Internal Memory Controller, have a maximum specified memory voltage (VDIMM) of 1.65V. You should bear this in mind when choosing DDR3 memory modules, as most (all? right now, I think it's all...) that are specced over DDR3-1600Mhz accomplish this using much higher voltage than 1.65V.

    Edit: http://www.anandtech.com/weblog/showpost.aspx?i=521

    This isn't what I was looking for, but it does contain some very interesting insight into i920 overclocking including memory settings.
  • edited December 2008
    Khaos wrote:
    Your memory choice won't hurt your CPU overclock. With memory multipliers all the way down to 6x, you will be fine with any DDR3 that is capable of 1200Mhz or so.

    What you do need to look at with you memory choice is the rated voltage:

    The Core i7 processors, due to their new Internal Memory Controller, have a maximum specified memory voltage (VDIMM) of 1.65V. You should bear this in mind when choosing DDR3 memory modules, as most (all? right now, I think it's all...) that are specced over DDR3-1600Mhz accomplish this using much higher voltage than 1.65V.

    Edit: http://www.anandtech.com/weblog/showpost.aspx?i=521

    This isn't what I was looking for, but it does contain some very interesting insight into i920 overclocking including memory settings.

    My problem isn't all the down, but all the way up. Not sure which max Uncore Multiplier as it limits max Memory Multiplier therefore limits max Memory frequency...

    Which if too low limits the max Bclock i can use which therefore limits the Max frequency on the CPU because max usable Multi in 920 is 20x.

    I hope this isn't too confusing.
  • KhaosKhaos New Hampshire
    edited December 2008
    You're thinking of this all backwards. Your memory clock / frequency does not limit your BCLK unless you've completely maxed it out, in which case you need to LOWER your memory multiplier in order to continue to increase your BCLK. If you RAISE your memory multiplier OR your uncore multiplier, that's going to mean you have to DECREASE your BCLK, which is NOT what you want to do!

    Here is an example that I hope makes sense:

    Pretend we have a memory stick with a max frequency of 36Mhz, ok? Like 1600Mhz, but smaller numbers.

    Say we also have a i920 that we KNOW can do 100Mhz, ok?

    We start with a BCLK of 2Mhz and the default memory multiplier of 8x


    BCLK = 2Mhz
    MMULT = 8x
    MCLK = 16Mhz .. OK, so we are well within spec for our 36Mhz memory, right?
    CMULT = 20x ...i920 Core mult
    CCLK = 40Mhz .... Our make-believe processor is running 40Mhz

    Now we increment BCLK by 1Mhz....

    BCLK = 3Mhz
    MMULT = 8x
    MCLK = 24Mhz .. OK, we are still within our memory spec!
    CMULT = 20x ...i920 Core mult
    CCLK = 60Mhz .... Goin' up! Overclocking is FUN!

    Now we increment BCLK by 1Mhz....

    BCLK = 4Mhz
    MMULT = 8x
    MCLK = 36Mhz .. memory is maxed! Now our BCLK is STUCK, right????
    CMULT = 20x ...i920 Core mult
    CCLK = 80Mhz .... Uh oh! We still have 20Mhz left in this processor!!!!!!!!!

    NO.

    BCLK = 5Mhz
    MMULT = 6x .. memory multiplier goes DOWN!!
    MCLK = 30Mhz .. memory is no longer maxed!
    CMULT = 20x ...i920 Core mult
    CCLK = 100Mhz .... Woohoo!

    500% OVERCLOCK ACHIEVED!!!! BUT WAIT...

    BCLK = 6Mhz
    MMULT = 6x
    MCLK = 36Mhz .. memory is maxed again!!! But now our CORE is running at 120Mhz instead of 80Mhz!
    CMULT = 20x ...i920 Core mult
    CCLK = 120Mhz .... Supar Woohoo!


    So Demon, this is the concept I was talking about when I said that your concerns are unfounded. If your cooling setup is good, you will be all the way down at 6x memory multiplier in order to achieve the highest possible BCLK and thus the highest possible CCLK.

    Please tell me this makes sense.
  • edited December 2008
    Well i have some things i have informed myself.

    Max Bclock around 200/210 depending on board/CPU

    Max Memory Divider i've seen is 12x

    To have 2000mhz in the memory we need 166 Bclock mininum.

    With 166 Bclock my max CPu frequency is 3324 approx.
    if i drop to 6x i need 333 bclock which it isn't possible.
    Even if there's an 8x i need 250 Bclock.

    CPU's are clocking around 3.8 to 4.2

    That's why i want to know whick is the max Uncore multi.
  • KhaosKhaos New Hampshire
    edited December 2008
    According to Intel all CPUs will have all memory dividers unlocked, and will have all the Uncore ratios unlocked, so you guys will be able to run even the highest memory speeds, even with the slowest Nehalem cpus.
    The article goes on the state that the maximum Uncore Multiplier is 30x for all of the processor models.

    Of course, there is no guarantee that any of these processors can run a 4Ghz Uncore in order to support DDR3-2000.

    But this is all beside the point.

    I am going to say this one more time and then I'm done:

    You would be so much better off getting memory that runs tighter timings at a lower frequency. If you don't understand what I mean by that, hit up google and read about the various timing parameters for DDR3 (Google: "CAS latency" or "DDR3 memory timings").

    Since the Core i7 has triple channel QPI bus, it provides huge memory bandwidth (i.e. sheer amount of data transfer) each second. What you're going to want to improve is the sub-timings that determine the actual number of cycles that it takes for read/erase/write to take place.

    I feel very strongly that this will yield the best perceived performance for the system, and the boys at Anandtech are backing me up. They are running their memory at something like 5-5-7-5-7 1N and only 1528Mhz (191 x 8) for their best performing i920 overclock with a CCLK of 4011Mhz (The i7's go into Turbo mode and add 1x to the CPU multiplier if you keep 'em nice and cool).
  • edited December 2008
    Is that the same article you posted above? 5-5-7-5-7 1N one.
  • KhaosKhaos New Hampshire
    edited December 2008
    5-5-7-5-7 1N one
    It's in the Anandtech "Lab Notes" blog entry I posted somewhere up there.

    Here it is:

    http://www.anandtech.com/weblog/showpost.aspx?i=521

    I have no idea where the Uncore multi quote came from. I just Googled "core i7 920 max uncore multiplier" and started reading.
  • DrLiamDrLiam British Columbia
    edited December 2008
    Great Explanations! I learned a lot :]
  • edited December 2008
    Khaos wrote:
    It's in the Anandtech "Lab Notes" blog entry I posted somewhere up there.

    Here it is:

    http://www.anandtech.com/weblog/showpost.aspx?i=521

    I have no idea where the Uncore multi quote came from. I just Googled "core i7 920 max uncore multiplier" and started reading.
    Actually its cas 7 at that speed, cas 5 is 1066 (which i doubt is better than 2000 cas9).
    Triple Channel Kits are coming out with cas 8 1600 at 1.65v and cas 9 2000 at 1.65v 6Gb kits.
  • KhaosKhaos New Hampshire
    edited December 2008
    My bad, you're right. I didn't look back at the article to double check.

    7-8-7-20 1N is their speed for 1528Mhz.

    You're probably going to have a lot of trouble running at 4Ghz uncore, but those cas 9 2000 kits might downclock well to faster timings. We won't know until some people write decent reviews of them.
  • edited December 2008
    Yeah, reviews wanted. ;)
Sign In or Register to comment.