Latency or MHZ, which is best?

Hello all

Im just posting this thread again in a different section because i think where i put it first might be limiting my repsonses ( i hope thats why no one has replied...) Sorry if you've read this already.

Hi everyone, can any help with this simple dilemma. I'm thinking of upping my RAM from 8GB of corsair CMX4GX3M2A1600C9 ( two sets of this) up to 16GB of one of these,
either CMX8GX3M2A2000C9 ( two sets) or CMX8GX3M2A1600C9. Both are on the approved list from the corsair website for my ASUS P7P55D motherboard with i5 chip.

The first has 2000mhz the second 1600mhz. The firsts latency is 9-10-9-27 the seconds 9-9-9-24. I dont understand the latency part and what is the bit to look for? lower latency numbers or higher MHZ.

Can anyone help me decide what ones to go for please and why.

Many thanks for reading and if you can help


I've just been reading some other threads to see if any match my dilemma, without any real luck yet and so i thought I'd add this info just in case it helps with any potential answers i may get. I'm running a lot of CAD stuff with lots of CPU and RAM intensive rendering so the extra RAM will help me. Also the RAM I'm using at the moment I've got running at the full 1600mhz by upping my BCLK clock in the BIOS up to 160. If i get the first set (2000mhz) then i'd probably try the same and up my BCLK to get that running at its full 2000MHZ if it likes it. I found that by turning up the BCLK to 160 it automatically OC'd my i5 from 2.75 up to 3.2 with out any problems.

Lastly my set up is water cooled (only the CPU) and have plenty of airflow so heat has never been a problem yet....

hope this info might help with any potential advice.

thanks again

27 answers Last reply
More about latency best
  1. I 'get' RAM, but first so I'm not confused -- if you already have 1 set of CMX4GX3M2A1600C9 then I would strongly suggest you get a second of Matching RAM. Otherwise the difference between {assuming 2 sets} CMX4GX3M2A1600C9 and CMX8GX3M2A2000C9 maybe as little as 1~3% gain AND you'd need to OC the CPU and your CPU that I don't see listed will also need to support the 2000MHz RAM; typically > 2.66 GHz.

    In general, for CAD I would NEVER recommend 2000MHz RAM. Also, I would recommend Xeon or X58 over any P55. My preference is Xeon with ECC RAM, {edit} and a Pro GPU rated for your CAD App.

    Please clarify your CPU and if you're wanting to Mix RAM?
  2. Thanks jaquith, heres what I currently have. Intel i5 2.75Ghz running at 3.2ghz after increasing my BCLK clock to 160 ( up from 133). I have two sets of CMX4GX3M2A1600C9 filling all 4 RAM slots that equals 4 x 2GB cards- total 8gb. I cant afford to start buying new processors and motherboards at the moment so i have to stick with what I have at the moment, but I can stretch to getting some more RAM. That looks like I'll have to pass on all 4 cards that I currently have installed ( i have friends who can use them and hopefully give me a couple of quid for them) and replacing them with 4 new ones ( 4 x 4GB) so that I can get the full 16GB that my motherboard will allow ( asus p7p55d).

    Can you tell me why you wouldnt recommend 2000Mhz for CAD/ what will it do or is it just because its not necessary. I checked out before I built my system what CPU would be best and my i5 or an i7 was recommended by the software company. Im using an nvidia quadro fx4500 x2 GPU, which seems OK to me, I got it cheap and it does the business. The render package I use is purely CPU dependant and so the GPU doesnt come into it ( as far as I've been led to believe

    Thats all the extra info I can offer I think, I hope that helps a bit.

    Im still confused about the latency though, is low latency more important than high MHz. I've just been assuming that higher MHz would mean faster RAM, I guess I'm wrong.

    Many Thanks

  3. 2000MHz is going to be producing a lot more errors, though the addition a BCLK of 160 does help reduce the disparity between the CPU Memory Controller and the RAM speeds. The error rate however when measured between the difference of 1600 MHz and 2000 MHz only is tolerable when 'gaming.' It is something that I would never recommend or build for your purpose. 2000 MHz 4GB sticks are okay for gaming.

    The Quadro FX 4500 X2 is as is any Quadro GPU a Pro GPU.

    For your purpose and what I know to be tested, I would recommend 2 Kits of Corsair Dominator CMP8GX3M2A1600C9, or possibly the Corsair Vengeance 8GB CMZ8GX3M2A1600C8 are newer but I'm not 100% certain that they have been tested for your MOBO. 4GB sticks need to be Tested or Certified before purchasing.

    In addition, you could get the 2000 MHz and run them slower; you'd need to contact Corsair for the CAS timings at 1600 MHz.

    If you were Gaming then I'd say fine go the 2000 MHz but only if you're OC'ing.
  4. Thanks very much Jaquith, im going to contact corsair and see if they have any tests or can recomend either of the ones your suggest. I'll post what they say as soon as i get a reply.


  5. Truly, at 1600 and 2000MHz the tiny differences in the timings of those two sets of RAM is moot and will make very little difference. For running CAD, the increased speed and going from 8GB to 16GB will be noticeable, again with the CAS making little difference.

    IMO, the CAS timings have become less relevant since RAM speeds have gotten above 1066MHz. Back in the day, with DDR2-800 going from CAS9 to CAS7 gave noticeable gains when running memory intensive apps, but as the speeds have increased, the differences in timings has become less an issue.

    All things being equal, if there was a set of DDR3-2000 with CAS12 and another with CAS9, then the CAS9 would be the better/faster set.
  6. thanks for the explanation malmental and chunky, that helped me a lot to understand things a bit better. So if you can treat me like a complete idiot then, have i got this right. If i have the choice of 2000mhz ram with C9 rating and higher latency or 1600Mhz with C8 rating and lower latency numbers, the 1600MHZ would be better as it will effectively move the processed info quicker.

    Thanks Rich
  7. I've lost count how many times I've posted the following information. Yes, Frequency and CAS timings {especially tight CAS timings 8-9-8-24 vs 9-9-9-24} all help produce 'faster' RAM processing, but the most important aspect is the IC's the RAM uses. You typically get what you pay for in most cases.

    Example of benched Speeds; the small gain from 1600 -> 2000MHz is not worth the errors and aggravation I've seen time and time again using 2000MHz RAM. Worst 2000MHz it's NOT always faster!!!

    Comparison Module #1: Mushkin Redline 996805 6-8-6-24 1600MHz
    Comparison Module #2: G.Skill Pi Series 7-8-7-24 1600MHz
    Comparison Module #3: Mushkin 998687 8-9-8-24 1866MHz
    Comparison Module #4: Corsair Dominator TR3X6G1600C8D 8-8-8-24 1600MHz
    Comparison Module #5: Kingston HyperX T1 9-9-9-27 1600MHz
    Comparison Module #6: Mushkin Blackline 998677B 7-7-7-20 1333MHz
    Comparison Module #7: Patriot Viper II Sector 7 9-9-9-27 1800MHz


    BTW - even though Bronze shows up I'm Gold; I answered so many posts correctly I broke the forum counter.
  8. Have to agree with jaquith on this one.
    (1) Overall performance is a funtion of both, some programs benifit more from the tighter timings while some benifit more from raw speed.
    (2) DDR3-1600 CL7 seems to be the sweet spot for i5/i7
    (3) Read a review that the bottom line was that DDR3-1600 @ CL 9 was about equal to DDR3-1333 @ CL 7. This comparision probably holds true for DDR3-2000/DDR3/1600. Author also indicated that CL8, or lower was best choice.
    (4) also agree with assesment on DDR3-2000 as higher speed even at the lower CL ratings lend to harder to obtain stability.

    Bottom Line - go for the DDR-1600 @ CL7
  9. Thanks once again Jaquith and to you too Chief. Its all becoming clear now what I need to look for. I'm pinning my hopes now on the vengance set you recommended Jaquith as it has the lower latency. I'm just waiting for corsair to reply to see if they have any tests to check it works OK. I wonder if there will ever be a truly plug and play society where everything is compatible with each other, things would be a lot easier..... thats life eh? lol

    cheers Rich
  10. I'm not a fan of XMP {Plug & Play}; I've had XMP either work perfectly fine <OR> BSOD/Post Failure even when it should work -- so I try XMP and if I get a problem then I quickly manually set the RAM per spec. Frequently enough, XMP won't work with 'some' 4 DIMM configurations particularly with Corsair.

    Therefore, my primary concern is Compatibility, and as I mentioned 4GB sticks are a particular concern. I know Corsair's website was down a good chunk of this morning {USA time}. Normally, they'll respond in a day or two, but with the holidays ... be patient.

    Let us know if the Vengeance work, I'm very interested. I do know the Dominator's I listed do work from a prior ticket I sent on your exact MOBO some time ago.
  11. just got this back from corsair

    If you want 16GB the best t use would be the CMZ16GX3M4A1600C9 or CMP16GX3M4A1333C9 16GB kits.

    In the last hour Ive been reading various other forums regarding my MOBO and RAM and I found someone who has recently put CMZ8GX3M2A1600C (9) onto an ASUS P7P55D E-PRO. They seem to have problem running it in XMP like you thought Jaquith. It works fine normally at 1333MHz but doesnt when they put the XMP mode on.

    It looks like the First suggestion from Corsair is is just two of the CMZ8GX3M2A1600C9 modules sold as one.

    whats your opinion on those two suggested Jaquith, their is little difference in price although the dominator set comes with a fan ( not that its any use to me as they dont fit onto my motherboard). I dont know what the difference is between dominator and vengance. They have the same latency but different MHZ.

    My thoughts are to get the vengance module but not use the XMP facility and try to OC it manually ( I might need your help with that too ;) ).

    Over to you, my RAM is in your hands...... no pressure :)

  12. I won't argue at all about 4X4GB Kits being the absolute best choice, they are indeed the best because they are ALL BIN Sorted together.

    CMZ8GX3M2A1600C9 OP 2X4
    CMZ8GX3M2A1600C8 JQ 2X4

    My vote is for the CMZ16GX3M4A1600C9 BIN Matched 4X4GB, you'll have the least problems and YES $10 is worth paying.

    Since your OC'ing the CPU I ALWAYS manually set the DRAM Voltage anyway, with a 4X4GB configuration you'll probably need to up the QPI/DRAM Voltage slightly. DO NOT get worried if the RAM does something unexpected after you install. If they're not damaged then a 5~10 minute tweak will have you up and running.

    Sure when you get the RAM just post here or Private Message me. Also, if you post your final choice now -- I'll post the changes needed in the BIOS; one as XMP {fingers crossed} and the second manually.

    Good Luck!
  13. Just ordered the CMZ16GX3M4A1600C9, it should arrive tomorrow apart from the fact that I ordered it from ebuyer, who have never got a previous order right and they use UPS who are always late ( i like to give people second chances and third and sometimes fourths....)

    I'd be really very grateful if you could post the figures for me, that would help me out big time. Thanks for all your help so far, its been very much appreciated.

  14. jaquith said:
    I've lost count how many times I've posted the following information. Yes, Frequency and CAS timings {especially tight CAS timings 8-9-8-24 vs 9-9-9-24} all help produce 'faster' RAM processing, but the most important aspect is the IC's the RAM uses. You typically get what you pay for in most cases.

    Example of benched Speeds; the small gain from 1600 -> 2000MHz is not worth the errors and aggravation I've seen time and time again using 2000MHz RAM. Worst 2000MHz it's NOT always faster!!!
    It's interesting seeing the comparison between the Mushkin Redline 996805 6-8-6-24 1600MHz and the G.Skill Pi Series 7-8-7-24 1600MHz. Given they are modules of the same speed with tiny differences in timings and the results shown in the graphs are, for all intents and purposes, almost identical. Aside from the larger numbers on a graph, I'd be surprised if there would be any noticeable real world benefit from choosing the module with the lesser timings.

    However, the results comparing the Mushkin 998687 8-9-8-24 1866MHz to the Corsair Dominator TR3X6G1600C8D 8-8-8-24 1600MHz are surprising. Just goes to show why Corsair remains the memory of choice.

    Good post jaquith, very informative. The OP definitely made the right choice in going with the Corsair CMZ16GX3M4A1600C9.
  15. Just for the heck of it I created a post -> ; my first post there. Since there's no/zip info out there on these NEW Vengeance I thought it will help many others.

    AI Overclocker Tuner -> D.O.C.P.
    DRAM OC Profile -> DDR3-1600MHz
    DRAM Timing Control {enter}
    DRAM CAS Latency -> 9
    DRAM RAS to CAS -> 9
    DRAM RAS to PRE -> 9
    DRAM RAS ACT -> 24
    DRAM Timing Mode -> 2N
    DRAM Voltage -> 1.5v
    IMC Voltage -> 1.2v~1.25v {try Auto first, generally maxing the DIMMs requires extra voltage - but you have low voltage RAM.}

    Lets see if the Corsair Gurus agree?!

    I assumed your CPU is the i5-760 (2.75GHz) and the MOBO the 'regular' ASUS P7P55D
  16. Thanks very much for posting that on the corsair site, I would have no idea of the right questions to be asking ( maybe one day if i keep learning these things ). You have assumed the correct CPU and MOBO, so hopefully any answers that come back from it will be good ones.

    just got to wait for the mail man now for my new purchase to try out the settings.

    ill keep you informed how it goes so that others can benefit too.
  17. The answer I got was AI Overclocker Tuner -> XMP. But, I also want the 'just in case' settings confirmed.
  18. Well the RAM turned up and is now installed, here's whats going on so far.

    First of all I turned it on with out making any changes to the BIOS ( accidently ) an error appeared that said CPU fan error press F1 to resume. I did that and it booted up fine. RAM was recognized and the CPU was still at 3216MHz.

    Then I rebooted and changed the settings to XMP, and carried on. the same error came up and when it was running the CPU had dropped to 2733 MHz

    lastly I rebooted and changed the settings to the ones you gave Jaquith. They actually turned out to be the same as what happens when its in XMP mode. Same error appeared and its still now runs at 2733 MHZ.

    Any ideas why the CPU fan error and why its back to 2733MHz.

    My system is watercooled with a Swiftech set up, what used to happen is the pump was controlled by the CPU fan and it changed speed with difference in temperature. Now it just runs at full speed, which is a little bit annoying and probably not good for it.

    something Ive noticed via speed fan, each core says 19 - 20 degrees ish, but the CPU temp reads 36. that seems high at least higher than it used to be.
  19. OK im stupid, I just checked in my case and I had dislodged the fan controller wire slightly so it was not quite home when I was swapping the RAM. At least that mystery is fixed..... Still would like to up the CPU speed to where it was before though if thats possible.
  20. Well I read the first post and I swallowed my heart, but then I read on...thanks above :)

    Q - I assumed that you raised the BCLK to 160? Keep in mind when there's no stress on the CPU it will lower the CPU Multiplier to conserve power. If you use both CPUz and Prime95 what are you getting then?

    CPUz -
    Prime95 -
    i5-760 {Multiplier = 21} (2.80GHz)

    BCLK 160 X 21 = 3200 = 3.36GHz ; stressed.

    Q - Does X.M.P. work 'now? I too assume my manual works now? I'd like to close the Corsair post and fill-in the blanks.
  21. how do you think i felt when i saw the error? :)

    I didnt need to raise the BCLK back as it stayed at 160 all the time.

    Im running prime now and CPUz, nothing is happening to the CPU speed, its reading 2731.2 with a multiplier of 17. if i look at the memory section the frequency is 803.3 MHZ and the FSB: DRAM is 2:10. Does that make any sense to you, it doesnt me.

    im going to reboot now and check what all the settings are at, as I left it in the XMP mode. In the bios it does state that XMP optimises everthing automatically and to OC via the BCLK it has to be done DOCP. maybe i'll try your settings again and see what happens.

    Ill report back soon.
  22. Ok just went back to BIOS and the CPU mulitplier was set to 17 so I put it up to 21 and now using CPUz its clocking at 3374MHZ. Running prime the temp of the CPU has risen from 36 idle up to 54.

    speed fan puts the MHZ at 3210 so im guessing its not as accurate as CPUz. It does seem to stay at 3374Mhz now even when its not stressing and just sitting.

    I also put it into your settings too and it works exactly the same as with the XMP mode so i guess all is fine.

    Have I done the right thing by turning the multiplier to 21? what happens if I go higher? or should I be happy with what Ive got now?

    In prime I ran the blend test, was that right or should I try the small fft or inplace large. Ive never really ran prime so not too sure what to do with it.
  23. sorry one more thing, i think im coming down with a cold so im off to bed now, so wont be able to try anything else until tomorrow now. Thanks for all the help so far, you're definately racking up some karma points :)
  24. I would have fallen out of my chair then gone to a dark room...and cried in a manly way.

    I'm confused that your BCLK is 160 without you manually changing it in the BIOS??!!

    Are you using ANY ASUS OC Utility then uninstall it! They are horrible and contradict or corrupt BIOS entries all of the time!

    It seems everything is going okay!
  25. Hello

    After a really bad nights sleep im back on it.just before i went to bed last night i ran prime 95 and i kept getting an error after test 5 or 6. I was running it in XMP mode and with a multiplier of 21. Today ive just changed my settings to the ones you gave me Jaquith and i set the multiplier to 20. The CPU is down to 3213MHZ from 3374MHZ with multiplier of 21. its now running prime and has got to self test 1024k passed, so its got further.

    Im going to leave it running for a while to see what happens. How long would you recommend?

    As for the ASUS OC utility i did have AI suite installed but was only using it to control the CPU fan speed. Ive uninstalled it now.

    Im running prime on the blend test, is that right or should i try another one
  26. Prime95 - to 'really' test do it overnight e.g. 8 hours. The last thing you want is to be knee deep into something {work/games} and poof!

    Yeah, don't use ANY Utilities {GA are the worst but ASUS is a close second} - great MOBOs bad software!
  27. Hello again

    Sorry ive not posted back with any results for a few days, my flu took hold and ive been in bed all week recovering. Ive finally been able to run prime 95 for a longer period than before, but strangely im getting fatal errors quite quickly. After about 4 tests on two cores it comes up with rounding was 0.5 expected less than 0.4. I have no idea if this means anything.

    Any ideas what I should try to fix this? should I adjust the BCLK down a bit ( currently on 160 ) or turn the multiplier back down ( currently on 20 ) as the XMP mode wants it to be on 17. If I keep on turn the multiplier back im losing the mhz gained on my CPU.

    PS Happy new year to everyone thats been helping with this thread and to everyone else that may have read it.
Ask a new question

Read More

Memory Latency