Announcement

Collapse
No announcement yet.

DDR5 RAM - Read Uncached Transfer Speed- Relevance to Practical Gaming Applications

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • DDR5 RAM - Read Uncached Transfer Speed- Relevance to Practical Gaming Applications

    Just shopping around DDR5 RAM for a new build, curious to know what the practical benefit of Read Uncached Transfer Speed is relative to real world gaming application (as opposed to productivity or intensive rendering workloads) and how the capacity of a 16GB stick (combined Dual channel 32GB) vs a 32GB stick (combined dual channel 64GB) impacts that performance category in particular. Generally speaking of course.

    I couldn't help but notice that, with all other factors being the same, a Dual Channel 16GB set at 6400 speed rating for example appears to consistently outperform a Dual Channel 32GB set at the same speed and CL rating - particularly in the Read Uncached Transfer Speed category. Same clock and CL but with larger capacity seems to take a hit in Read Uncached Transfer Speed - is that fair to assume overall?

    1) Can anyone help enlighten me as to why this is and perhaps more importantly, what the trade off is here in terms of gaming performance (or how relevant that category is to gaming specifically)?
    2) Is it better to sacrifice capacity (32GB total vs 64GB) in favor of better Read Uncached Transfer Speed with gaming performance in mind (understanding most games won't ever need beyond 32GB capacity anyway)?

    I appreciate your time and any thoughts anyone might have on this, thanks

  • #2
    There are lots of things that can have a minor impact. Channels, ranks, timings, NUMA, x8 or x16, bus speed, CPU capabilities, etc...

    Some high level points.
    1. Generally speaking more RAM does not equate to faster performance. Quantity does not equal speed, unless you have run out completely.
    2. Single channel is bad. So two matching sticks are significantly better. Quad channel is better again.
    3. XMP is good if available
    4. Super high end RAM is a waste of money. Mid range is fine.
    5. CPU caching hides a lot of the impact of slower RAM.
    6. Once you have "enough" mid range RAM, a matched pair and turned on XMP, anything else you will only have a minor impact. So minor you'll likely never notice.

    This table is a nice example of real world numbers

    See table below for a comparison of the Read, Write, Latency and Threaded results when running XMP Profiles.
    RAM Test 1 Stick
    Default Settings
    Single Channel
    2 Sticks
    Default Settings
    Dual Channel
    2 Sticks
    XMP Profile 1
    Dual Channel
    2 Sticks
    XMP Profile 2
    Dual Channel
    4 Sticks
    XMP Profile 2
    Quad Channel
    Percentage Improvement
    Worst vs Best Result
    Read (MB/sec, 1 CPU core) 17161 17416 17808 18551 18871 9.9%
    Write (MB/sec, 1 CPU core) 7983 8351 8397 8512 9056 13.4%
    Latency (nano sec)* 43 43 43 43 42 2.3%
    Threaded (MB/sec) 19923 27499 27854 28809 55027 176%
    CPU Mark 9055 10822 10919 11616 13407 48%
    *Lower values are better.

    Comment


    • #3
      Thanks very much for the thoughtful response, I appreciate this perspective from a relative expert.

      This general guideline will go a long way to prevent me from agonizing over minute details that ultimately won't impact my day to day use-case. Thanks again!

      Comment

      Working...
      X