RAM Speed/Latency ...what effects to expect

Discussion in 'Tech Talk' started by Krazie_Ivan, Jun 30, 2018.

  1. recently in another thread, we had some debate as to if it's worth spending slightly more for faster/better memory (higher MHz & lower CAS).

    quite a few variables determine if it's going to be of value to upgrade; while it's mostly down to the particular game or program code, some architectures will also respond with higher margins compared to others... Ryzen is a good example of this, but all Intel chips have varied results too (albeit typically less pronounced). another factor is where your planned system components will fall in terms of overall performance - the higher-end it is, the less your RAM selection has an impact on your total cost ratio. if you are planning a top-tier CPU & GPU, then the only two ways to further boost gaming & app performance is going to be your RAM and to an even lesser extent, an SSD.

    a basic rule-of-thumb when looking to buy modern DDR is: take the speed (lets use 2400Mhz as an example), remove the zeros (24), divide by 2 (12)... that is the CAS/CL number you want to look for, and any lower is even better. if the CAS/CL is higher (w/o a corresponding increase in MHz), it's slow RAM & not a very good match for a performance PC build. the CPU waits extra clock cycles to get info from the memory, so performance is considered "good" when True Latency calculates out to less than 10ns (clock cycle time X number of clock cycles = true latency).

    when you start shopping, there comes a point where the added cost of increasingly faster RAM becomes a poor value proposition, but that point is always in-flux with prices. so it's a good idea to keep a ratio of performance/dollar in-mind, and go with what's on sale at the time that makes sense. while the benefits can be different for each game or app, it's worth remembering that you'll likely not replace your RAM every time you play a new game or start using a new program ("i play ___, which doesn't care how fast/slow my RAM is.", but the next one may).

    overclocking & tightening latency timings provide "free" performance boosts... but typically require higher quality RAM die modules, a higher-end mobo with clean power delivery, quality PSU for stability, & a bit better cooling is always helpful. while small bumps in MHz are pretty easy for peeps less-experienced, manually setting tight sub-timings takes more time/effort/knowledge/patience. for the games/apps that respond well, adjusting your memory settings can be well-worth the effort... as you'll see below.

    for members looking to upgrade in the future, i thought it'd be helpful to provide a basic lil cheat-sheet list of how certain games respond to faster RAM (higher MHz, lower Latency, tightened sub-timings). Average framerates are important, but 1% and .1% lows are an indication of stuttering... which can translate to a really poor experience even if the Avg is far higher.

    - BF1 : yes, especially sub-timings
    - Titanfall 2 : no, only .1% lows improve
    - Rise of the Tomb Raider : yes, especially sub-timings
    - GTAV : moderate improvements
    - Hitman '16 : yes, but only till 3200
    - Dirt 4 : yes to avg framerate, but not much to 1% and .1% lows
    - F1 '17 : yes, but not so much to 1% and .1% lows
    - Project Cars 2 : yes, but only up to 2933, although sub-timings play significant role
    - Forza Horizon 3 : yes
    - Ashes of the Singularity : yes, especially sub-timings and 1% lows
    - Gears of War 4 : yes, but minor incremental boosts above 2666
    - The Witcher 3 : yes, especially .1% lows
    - Civilization 6 : no (time to turn, not FPS)
    - Middle Earth, Shadow of War : very small incremental boosts
    - Far Cry 5 : very small incremental boosts
    - Kingdom Come Deliverance : very small incremental boosts above 2400
    - Assasin's Creed Origins : small incremental till 3200, sub-timings are just as important
    - Elder Scrolls Online : very small until 3200, sub-timings very important to 1% lows
    - Fortnite : yes, but game performs insanely well at any speed ....1% lows get biggest boost
    - CSGO : yes up to 3200, but game performs insanely well at any speed

    Average FPS Boost
    2133/2400 = 7.5%
    2400/2666 = 4.4%
    2666/2933 = 3.4%
    2933/3200 = 4.6%
    3200/3333 = 1.7%
    3333/aggressive sub-timings = 7%

    source of Ryzen 1 testing:

    i could only find a good benchmark methodology vid with Ryzen (1800X, from June 2018), as the Intel testing i've found so far have too many moving variables or flaws to provide accurate/dependable results (but i will keep looking & edit this post when i find a good source for Coffee & Ryzen 2xxx).
    #1 Krazie_Ivan, Jun 30, 2018
    Last edited: Jul 7, 2018
  2. Your calculation for CAS Latency (take the speed (2400), remove the zeros (24), divide by 2 (12)) would be great but getting to that point just isn't how the RAM is scaling. As speeds for RAM increases, so does the CL. We may want those CL's but actually getting those timings aren't something that you'll be getting from a stock kit without paying a huge premium(Corsair Dominator Platinum kit is $215, 2400mhz CL10). If you spend the time to research and overclock on your own, those timings MAY be achievable, though its highly dependent on which mobo and RAM manufacturer you have. Here's the example that Crucial gave in their write up.


    If you're interested in the super in depth workings of DRAM, check this video out. It's a bit less "bubblegum" :p

    Here's their written article if you don't want to watch a video.


    With Ryzen, RAM speed is a huge deal to get the most performance out of your system, with Intel, not so much. Most RAM kits on the market will overclock without issue. There is plenty of headroom when it comes to voltages. I would spend time researching the specific OC you want for the motherboard you have. If you're willing to put the time in to OC your CPU, spend a bit more time and do the RAM as well.

    In a previous post, I mentioned that when it comes to price/performance, RAM speed isn't where you should spend your money. I still stand by that, unless RAM prices drop to 2016 prices(sub $100 for 16GB's). Should that happen, go ahead and spend the extra as its a few dollars instead of $20-50. For general gaming, I suggest buying the most amount of RAM possible for the cheapest price. 8GB's of RAM while playing BF1 is barely enough these days, so 16GB's for gaming is now the standard.
  3. maybe i wasn't clear enough in my post (?), but the calculation is not for finding c12 RAM (2400mhz & c12 were only used as an example). the calculation is only solving by ratio to find an acceptable baseline performance of about 10ns True Latency (which is NOT the same as rated CAS Latency numbers). because it's simply a ratio, it scales & can be applied to any speed:
    2133 = c11
    2400 = c12
    2666 = c13
    2800 = c14
    3000 = c15
    3200 = c16
    3400 = c17
    3600 = c18
    3800 = c19
    4000 = c20
    and so-on
    ...so, if you're looking at purchasing memory at any factory spec'd advertised stock speed & want to know if the advertised latency is better or worse than ideal, apply the calculation & see where it falls. if you're shopping 3200mhz RAM for your performance system build & one says c15, the other says c17 ...buy the c15 kit if the price difference is nominal & all other factors are the same (reviews/features/support/etc).

    thx for the feedback - i def don't want to cause confusion, so i'm glad to explain further (and edit for clarity). GN did really well with that article & vid. very thorough with explanations of an extremely detailed topic.