No announcement yet.

RAM benchmarks for less than optimal configurations

  • Filter
  • Time
  • Show
Clear All
new posts

  • RAM benchmarks for less than optimal configurations

    We see a lot of RAM related performance problems. So we thought it was worth quantifying how much of an impact the RAM configuration can have on the benchmark results.

    The two main issues are:

    Accidentally running in single channel
    Most motherboards offer dual, tri or quad channel support. RAM access can be interleaved in dual channel mode. So they work faster with 2 or more sticks, but only when the sticks are in the correct slots. It is a common problem with cheap machines that they ship with just 1 RAM stick, preventing dual channel mode from working. So check the motherboard manual for the optimal slot usage.

    BIOS setup
    Another common problem is the BIOS configuration for the RAM. Vendors often select a safe (slow) default settings for timings and clock speeds. There is often a setting in BIOS to select a faster "XMP profile". A XMP profile is a collection of faster settings that should be safe to use, as they were validated by the RAM vendor. For advanced users who don't mind of bit of instability you can manually set the clock speeds, timings, voltages & gear selection. Generally using one of the XMP profiles is sufficient however.

    Finally make sure you have enough RAM to avoid running out and forcing the operating system to swap memory pages out to disk. 8GB is really a bare minimum for Windows 11. Not having enough RAM can be devastatingly bad for performance, but once you have 'enough' adding more doesn't improve performance. (More RAM does not equal fast RAM).

    RAM benchmark results with PerformanceTest using different RAM setups on the same machine (DDR4 RAM).
    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
    Percentage improvement
    Worst vs Best result
    Read (MB/sec, 1 CPU core) 17161 17416 17808 18551 8.1%
    Write (MB/sec, 1 CPU core) 7983 8351 8397 8512 6.6%
    Latency (nano sec) 43 43 43 43 0%
    Multi-Threaded (MB/sec) 19923 27499 27854 28809 44%

    This is just an indicative example. Different CPUs and different RAM will give different results.
    Another interesting element in the results is that 1 CPU core doesn't max out the RAM's bandwidth.