PassMark Software

Announcement

Collapse
No announcement yet.

memory latency inconsistencies

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

  • memory latency inconsistencies

    Hi the one line summary of what I'm looking for is "given CPU X, how do I pick out the lowest latency DDR4 memory?" I was hoping it would be as simple as choosing the part that had the lowest latency on the benchmark page, but it doesn't seem that simple.

    More detailed question:

    I was hoping someone could explain the discrepancy between the DDR4 Intel memory latency benchmarks I see, and what one would expect the latency to be based on a CAS latency lookup table found for example on wikipedia:

    https://en.wikipedia.org/wiki/CAS_latency

    Benchmark link for handy reference:

    http://www.memorybenchmark.net/latency_ddr4_intel.html

    If you lookup DDR4-3200 @ 15 cas latency, first word is 9.38ns.
    If you lookup DDR4-3200 @ 14 cas latency, first word is 8.75ns.

    An equivalent part on the benchmark towards the top is F4-3200C15-8GTZSK 8GB which shows 17ns.

    Another part further down is F4-3200C14-8GTZSK 8GB which shows 22ns.

    Why is a cas 14 latency part showing 23% more latency than a cas 15 latency part, with the same size, same manufacturer, same 3200MHZ part. If anything I would expect the cas 14 latency to always perform better at the same frequency. Is this because they were tested in different CPUs or something like that? If so is there any way to align the benchmark results uniformly to get truly the best latency parts keeping the CPU/motherboard constant?

    Alternatively is there at least a way to drill down and see what platform this benchmark came from so I can try and gauge equivalence manually?



  • #2
    The underlying problem is that many (even most) CPUs can't make full use of the highest speed RAM. Even in cases where the CPU should be able to fully use the RAM, many people never go into BIOS and set the correct XMP profile.

    The latency numbers we measure aren't a direct measure of the CAS. But low CAS should also give a lower latency benchmark if all else is equal (and the CPU is up to the task).

    You can look up individual baseline files from within the PerformanceTest software.

    Comment

    Working...
    X