Advertisement
If you have a new account but are having problems posting or verifying your account, please email us on hello@boards.ie for help. Thanks :)
Hello all! Please ensure that you are posting a new thread or question in the appropriate forum. The Feedback forum is overwhelmed with questions that are having to be moved elsewhere. If you need help to verify your account contact hello@boards.ie

RAM: faster MHz or lower CL

Options
  • 07-12-2008 1:47pm
    #1
    Banned (with Prison Access) Posts: 16,659 ✭✭✭✭


    I always get confused by this: when I go to order memory from the likes of crucial I'm usually offered two options, one with a higher MHz and one with a lower CL. Example attached. Which should I be going for?

    Thanks,
    adam


Comments

  • Registered Users Posts: 17,435 ✭✭✭✭Blazer


    The CL is the latency for memory.
    The lower the CL the faster memory transfer happens.
    In the screenshot the first kit is better than the 2nd for latency however the 2nd kit has a faster memory frequency.
    Ppl has various opinions on latency and speed. Normally the higher the frequency the higher the latency.

    Read the link below for more info on ram frequency and memory timings (cl)

    http://www.thetechrepository.com/showthread.php?t=160


  • Registered Users Posts: 5,560 ✭✭✭Slutmonkey57b


    Anandtech is of the opinion both speed and latency are less important on DDR kits than tRd ratios:
    http://www.anandtech.com/showdoc.aspx?i=3208&p=7
    Long article but the numbers on the front page speak for themselves.


  • Banned (with Prison Access) Posts: 16,659 ✭✭✭✭dahamsta


    Thanks lads, but my question hasn't really been answered. I understand that a lower CL is better and a higher clock rate is better, but if both are offered, which is the better option?

    I don't think trd info is available. That article is way over my head. :)

    adam


  • Registered Users Posts: 1,186 ✭✭✭Nichololas


    I was going to ask this soon as I've building a new computer over the break and I was wondering what the difference is between ram timings/bandwidth, so I might as well ask it here ..

    Example

    4096MB-Kit OCZ DDR2 PC2-8500 Gold Dual Channel, CL5
    versus
    4096MB-Kit OCZ DDR2 PC2-6400 Reaper, CL4

    Is there a general formula that can be applied to these to determine which is better?


  • Moderators, Technology & Internet Moderators Posts: 18,377 Mod ✭✭✭✭Solitaire


    Go for the module in your price range that has the best combination of high clock speed AND low CL rating. And factor in the rated voltage that it is supposed to work at - lower is better, and having a gap between rated and safe maximum voltage given is also a sign of additional OC headroom (so long as your mobo can handle the maximum voltage in question!) I've seen cheap PC5300 kits rated for CL4@1.8v overclock far past basic PC6400 rated for CL5@1.9v so long as you drop the CL on the first up to CL5 - at CL5 and 1.9v you can go way past 900MHz on the "slower" modules! And there is probably a similar issue with high-grade PC6400 CL4 beating power-hungry PC8500 CL5 if you mess with it :)

    And if you can just drop the tRD (aka "Performance Level") down if you can after you've decided what speeds to run your FSB and RAM at. The default is 12-15 but can be as high as 21 - all of which are WAY too high! Higher memory speeds, lower FSB and a greater memory divider ratio lends to lower tRD and better performance - as a rule of thumb, say the lowest feasible is tRD4 and most low-mid-end systems can survive with tRD dropped all the way down to 5; systems with high FSB (400+) should use a tRD of 6 or even 7 or 8 for high-FSB, low-divider systems (again, interrelated; massive FSB may increase performance but it forces tRD higher which reduces bandwidth. Swings and roundabouts!) Just experiment a bit to see what works out.

    Edit: Sorry Nick, you posted after I started my post. Both are very close in spec, the Reaper might run cooler but the teeny fin array might block some CPU coolers. Both run around the same voltage range, the Gold seems a bit nicer but bear in mind its only just CL5, most of the ratings seem to be closer to CL6 timings! You might want to see some reviews for the two...


  • Advertisement
  • Banned (with Prison Access) Posts: 16,659 ✭✭✭✭dahamsta


    Solitaire wrote: »
    Go for the module in your price range that has the best combination of high clock speed AND low CL rating.
    I don't have that option. As I explained in posts #1 and #4, it's an either/or situation. See the image in the first post.


  • Moderators, Technology & Internet Moderators Posts: 18,377 Mod ✭✭✭✭Solitaire


    Didn't see the ECC first time around.

    Cock.

    Unless its VERY specific or you are getting some kind of deal on it a possible alternative is this Kingston module - buy 2 and stick them in in dual-channel and even with p&p it probably works out cheaper than the crucial, unless you have some kind of big offer on them. Those are PC6400 AND CL5 at the same time :P

    If you have no choice other than the original modules shown I'd say go for the PC6400 (800MHz) kit. Same voltage, slight difference in latency, should give a slight edge over the PC5300. But I doubt its a very big margin at all - again, check what serves the intended application(s) best, higher clocks or lower latencies.


  • Registered Users Posts: 5,560 ✭✭✭Slutmonkey57b


    I don't have that option. As I explained in posts #1 and #4, it's an either/or situation. See the image in the first post.
    Thanks lads, but my question hasn't really been answered. I understand that a lower CL is better and a higher clock rate is better, but if both are offered, which is the better option?

    I don't think trd info is available. That article is way over my head. :)

    adam

    The reason you haven't got an answer is that it's not that simple - it depends on what are doing! And the gist of the article above is this:
    Speed ratings and CL ratings are far *less* important than tRd when it comes to performance improvements. So you could stick out a whole wad of money on high speed rated, low CL RAM, but some other guy that's fiddled with his tRd setting on a lower rated, cheap ram with slower settings on FSB and CL could be getting better results.

    The only reason to care about speed vs. latency is if you're likely to overclock. If you require ECC registered RAM for a server, then you're looking for a different sort of advice altogether.

    Most intel processors have an FSB of 266-333 Mhz "quadpumped" to the rated value of the processor (so it's advertised as 1333Mhz, as an example). DDR ram is rated to cover FSB of 233-600 "Double Data Rate", so it's advertised as 800.

    So for example an e8400 has a "rated" FSB of 1333Mhz - or 333Mhz "real".
    DDR2-5400 (or DDR2-667 as it might be called) also runs at 333Mhz.

    If you are going to overclock, then you are raising the FSB, and you need RAM that will run at a higher speed. This means you either attempt to run your RAM at a higher rated speed than it claims to be able to do, which may involve bumping up voltages and/or loosening CL timings, or buying a higher speed RAM which will the manufacturer guarantees will run at that speed.

    If you are a novice overclocker, and using an Intel chip, then most DDR-800 kits will keep pace just fine. If you're overclocking an AMD, then it's a lot more fiddly. If you're not overclocking, then just buy the cheapest RAM you can get that's compatible with your chip's FSB, and get more of it.


  • Registered Users Posts: 5,560 ✭✭✭Slutmonkey57b


    Enlil_Nick wrote: »
    I was going to ask this soon as I've building a new computer over the break and I was wondering what the difference is between ram timings/bandwidth, so I might as well ask it here ..

    Example

    4096MB-Kit OCZ DDR2 PC2-8500 Gold Dual Channel, CL5
    versus
    4096MB-Kit OCZ DDR2 PC2-6400 Reaper, CL4

    Is there a general formula that can be applied to these to determine which is better?

    Like I pointed out in the article, knowing where to find tRd settings is going to give you more of a boost than worrying about rated FSB's and CL timings. CL was very important for the last generation, but less so for this one (eg DDR3 runs at huge CL compared to DDR2 but outperforms it in many cases)
    Unless you're prepared to fiddle in your bios a lot, and you're confident about fiddling with voltage levels, the average user doesn't need to worry about the difference between 8500/CL5 vs 6400/CL4. If you know what you're doing, 6400 RAM can perform just as well as higher rated modules. If you don't know what you're doing, you're better off leaving it alone.


  • Banned (with Prison Access) Posts: 16,659 ✭✭✭✭dahamsta


    Thanks Slutmonkey. No overclocking and they're the same price, so I guess I'll just flip a coin. TRD still means nothing to me and tbh I've no interest in reading an article that refers to it constantly without ever explaining it. Don't see it on Wikipedia either. Got a one-line explanation? :)

    Thanks all for your help.

    adam


  • Advertisement
  • Moderators, Technology & Internet Moderators Posts: 18,377 Mod ✭✭✭✭Solitaire


    Yeah, as in my last post just ignore all the OC stuff and go for the Kingston if you can and the PC6400 Crucial if you can't and you know your server can take 800MHz RAM. If you want to play it safe just get the PC5300 - the speed difference isn't very big at all and it has maximum compatibility with older DDR2 mobos.

    As for the one line explanation...

    tRD is the number of memory clock cycles that the memory controller (MCH) will wait for data that's already been written into the RAM to travel back across the MCH to the CPU to be read by it (thus the term "Read Delay").

    EDIT: Damn. Two-line explanation :p

    AMD and the new Corei7 have their MCH inside the CPU, speeding things up. Older Intel CPUs have to rely on a seperate MCH on the motherboard, slowing down the speed at which the CPU can find out what's been written in memory. The less time you give the MCH to get the data to where its going the better, but of course if you don't give it enough time you end up trying to read data that hasn't even arrived yet and it all comes apart (well, crashes) :eek: The problem stems from the fact that Intel has been erring on the side of caution to a stupendous extreme, slowing things down unneccesarily with a slightly absurd one-shoe-doesn't-fit-all-but-we'll-try-anyway strategy. Which leaves it to the end user to tweak tRD to suit their system. If they even can. Nice one Intel :p


  • Banned (with Prison Access) Posts: 16,659 ✭✭✭✭dahamsta


    Thanks very much Solitaire. Why Anand couldn't have done that I don't know!

    adam


Advertisement