1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Anyone know what the bandwidth-based performance limit for Ryzen might be...?

Discussion in 'Processors and motherboards AMD' started by waltc3, Mar 12, 2018.

  1. waltc3

    waltc3 Master Guru

    Messages:
    631
    Likes Received:
    60
    GPU:
    Asus RX-480 8GB
    Now that my ram is running at 1600Mhz (3200) with stock settings, I decided to briefly experiment with overclocking my overclocked ram...;) I am currently running fine @ 1633MHz (3266); and of course I tried 1666.5 Mhz (3333), and was surprised to see that it is actually 98% stable and booting was no problem--thanks to AMD's latest 1.1.0.x AGESA. (But 98% still meant the occasional GSOD, so I have throttled back to 1633Mhz with 100% stability so far.)

    Anyway, the performance I am talking about is not related to overclocking the cpu as much as it is what happens with Ryzen when ram speed/bandwidth increases. Running at 3.8GHz on the cpu (my "stock" setting for my R5 1600), and running several cpu benches, with each iteration above 1600MHz (3200), the results ratchet up higher, in line with the increased bandwidth.

    So I'm wondering what the limit for the current Ryzen architecture might be in terms of performance increases gained through DDR4 bandwidth increases (with a static cpu clock)...I would expect that at some point the cpu performance gained through DDR4 bandwidth oc'ing would peak and remain static, regardless of how much higher the DD4 ram is clocked. It's just that I haven't found that ceiling yet--3333Mhz on the sys ram isn't the ceiling and I seriously doubt that 1700MHz (3400) would be, either. It looks like the only thing holding Ryzen back atm is available DDR4 bandwidth. Anyone run their own experiments?
     
    jaggerwild likes this.
  2. GroinShooter

    GroinShooter Master Guru

    Messages:
    442
    Likes Received:
    98
    GPU:
    GTX 1080 SC
    I've been under the impression that it's in the 3400-3800ish range, could be wrong though.
     
  3. Clouseau

    Clouseau Ancient Guru

    Messages:
    2,039
    Likes Received:
    102
    GPU:
    ASUS STRIX GTX 1080
    Have not seen anything beyond 3733 that passed HCI memtest. Personally not able to get 3500 stable with voltages am comfortable with but 3487 is good to go. Would agree with GroinShooter's estimate.
     
  4. AsiJu

    AsiJu Ancient Guru

    Messages:
    4,866
    Likes Received:
    605
    GPU:
    EVGA 980Ti SC+
    Memory speed wise the 3500ish range is the current limit but if you meant how far Ryzen could scale with mem speed then that's a good question.

    I saw a chart of Kaby Lake CPU scaling with memory speed (forgot where and about what exactly) and it was interesting, past 3000 up to about 3600ish there were diminishing returns IIRC.
    But at and past 4000 perf. scaling picked up again. KL got a notable benefit for past 4000 speed.
    In that particular scenario.

    Begs the question what the chart would look like for Ryzen past 4000 memory speed...

    really hope Zen+ and 400 series chipsets (at least) can break 4000 mem speed. And then some for Zen 2!
     
    Last edited: Mar 15, 2018

  5. phatbx133

    phatbx133 Master Guru

    Messages:
    805
    Likes Received:
    2
    GPU:
    MSI GTX 1050ti 4GB
  6. -Tj-

    -Tj- Ancient Guru

    Messages:
    15,084
    Likes Received:
    590
    GPU:
    Zotac GTX980Ti OC

    Here they show up to 7-10fps difference 3000 vs 4000MHz
    https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html

    While here only 1-3fps or negative
    https://www.techpowerup.com/reviews..._Memory_Performance_Benchmark_Analysis/9.html

    Maybe tpu had too loose tertiary timings.
     
    AsiJu likes this.
  7. AsiJu

    AsiJu Ancient Guru

    Messages:
    4,866
    Likes Received:
    605
    GPU:
    EVGA 980Ti SC+
    Thanks, yeah definitely makes a difference. And those articles were up to 4000 w/ Skylake and Coffee Lake.

    The one I read IIRC was about Kaby Lake and showed memory speeds past 4000. The performance scaling upwards from 4000 was better than with say between 3600 and 4000.

    I'll link the article if I'm able to find it again. Needless to say KL clocked at 5 GHz with 4000+ speed RAM was pretty impressive to look at.

    Worth noting though as TPU's article shows that latency matters a lot too.
    However w/ Ryzen it's somewhat established that push MHz as high as you can even if that means increasing latency as the infinity fabric works at the speed your RAM does.

    So the higher you can set your RAM clock the faster your CPU is, even if that may actually lower RAM throughput via increased latency.
    Practical benefit then ofc depends on whether the increased CPU performance can surpass the lowered RAM performance and depends on application.

    For example I'm running my RAM @ 2933 CL14 instead of 3200 CL16 (which it could do) as the difference in performance even in benchmarks is barely noticeable.
    Cinebench gave a couple of points higher score w/ 3200 than 2933.
     
    -Tj- likes this.

Share This Page