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

Reviews AMD Ryzen 9 3950X Review

Discussion in 'Article Discussion' started by bit-tech, 14 Nov 2019.

  1. Sn3akr

    Sn3akr What's a Dremel?

    Joined:
    26 Feb 2011
    Posts:
    51
    Likes Received:
    15
    Why was this tested with 3400 MHz RAM? It should perform better with even multipliers to the infinity fabric.. Gamers Nexus has a video on this, they found that 3866 run @3800 cl 16, with a 1900 clock on the infinity fabric gave up to16% higher fps in games compared to the worst of their tested. So for rysen 3 i'd recommend to get lowest cl possible with RAM divisable with 1800 (no OC) or 1900 (OC infinity fabric) as this can give a not insignificant boost to performance.
    See the video here.
    https://www.google.com/url?sa=t&source=web&rct=j&url=#&ved=2ahUKEwj5s96fsPjlAhXNs4sKHfS-ChIQwqsBMAp6BAgGEAk&usg=AOvVaw0VTJzWxrN8ZFOD4xbU2nov
     
  2. GeorgeK

    GeorgeK Swinging the banhammer Super Moderator

    Joined:
    18 Sep 2010
    Posts:
    8,649
    Likes Received:
    478
    No doubt this CPU would perform even better with better RAM but for the results to be comparable, the same RAM has to be used for each tested CPU.
     
    MLyons likes this.
  3. Dogbert666

    Dogbert666 *Fewer Staff Administrator

    Joined:
    17 Jan 2010
    Posts:
    1,678
    Likes Received:
    181
    ^ This. We could have taken other approaches, such as the maximum "officially" supported speed being used for each one, or finding the maximum stable speed for each CPU, but the former is an unrealistic use case for enthusiasts and the latter is too time-consuming. 3,466 is the highest stable speed across the board, and we're talking a lot of CPUs there!
     
    MLyons likes this.
  4. sandys

    sandys Multimodder

    Joined:
    26 Mar 2006
    Posts:
    3,874
    Likes Received:
    290
    Strictly speaking the machines should be tested in their official supported state and any other speeds should be in the overclocking section.

    I always find it quite odd that a review site like bit tech would massively overclock one systems RAM versus another it's not like using one speed is in the aid of leveling the playing field here, a 9900k should only have 2666Mhz Ram and Ryzen 3200Mhz.

    Once you start OCing RAM there are so many other variables like timing, IF clks etc. Of course to apply an equal memory overclock to a Ryzen system would require RAM over 4100Mhz which is probably not so straight forward even on Gen 3.
     
    Last edited: 20 Nov 2019
    Corky42 likes this.
  5. Spraduke

    Spraduke Lurker

    Joined:
    23 Sep 2009
    Posts:
    657
    Likes Received:
    141
    I hadn't twigged that the results are using overclocked memory (yeah yeah I should read the articles fully!).

    I echo Sandys thoughts that tests should be performed using the stock memory speed for non-OC to give a fair representation of performance. The OC result can then be at the fastest memory and CPU frequency achieved for that chip.

    This would give you a fairer idea of performance gain from overclocking and/or buying faster memory. It does seem unfair to 'slow down' memory speeds on one CPU just because another cant run that speed - that's another differentiator between the two CPUs I would like to know about.
     
  6. edzieba

    edzieba Virtual Realist

    Joined:
    14 Jan 2009
    Posts:
    3,909
    Likes Received:
    591
    If you really want to complain about RAM 'overclocks' then just give everyone JEDEC speed: 2133MHz.
     
  7. Corky42

    Corky42 Where's walle?

    Joined:
    30 Oct 2012
    Posts:
    9,648
    Likes Received:
    386
    When JEDEC start making CPUs then sure. ;)
     
  8. sandys

    sandys Multimodder

    Joined:
    26 Mar 2006
    Posts:
    3,874
    Likes Received:
    290
    It's not a complaint, it's an observation, the review is testing performance against competing parts, the two CPU architectures have officially supported RAM speeds that are higher than 2133 out of the box, if you are to compare them stock then they should be stock.

    Not one with a 30% RAM overclock vs the other with an 8% overclock particularly as the way the machines are tested here in low res and encodes etc are where the reduced latency and increased bandwidth that comes from good memory performance improves the scores fairly significantly.

    It doesn't matter to me a great deal because I do spend the time maximizing my system for greater performance but overclocks such as this can make one CPU look so much stronger than it is in reality, it is good for people to understand this particularly when the myth perpetuated is that memory speed does not matter.

    Example of memory speed on Intel platform can be seen here.

    https://www.tomshardware.com/uk/reviews/best-ram-speed,5951-6.html
     
  9. edzieba

    edzieba Virtual Realist

    Joined:
    14 Jan 2009
    Posts:
    3,909
    Likes Received:
    591
    They're both getting tested with RAM running at the same speed. Heck, they're using the same RAM for all tests. That's about as fair as you can get.

    After all, RAM prices do not change based on what CPU you plan to use them with. If one CPU has performance reduced by cheaper RAM and one doesn't, that's an important factor in a purchasing decision.
     
  10. Dogbert666

    Dogbert666 *Fewer Staff Administrator

    Joined:
    17 Jan 2010
    Posts:
    1,678
    Likes Received:
    181
    The debate here is a perfect illustration of there sometimes being multiple approaches to fair testing. To be clear, we consider both methods (highest official vs. equal across the board) to be fair, but we had to make a call one way or another. With our method, we get complaints that there is bias one way, "you're applying an overclock!" even though anyone reading these pages is likely to do exactly the same; with the other it swings the other way e.g. "you guys are so dumb, no one uses Intel CPUs with 2666MHz RAM, why are you artificially limiting them, how much did AMD pay you?" etc.

    The main thing is to have a sufficiently justified method and to be clear about it so that users are informed about what the testing does and doesn't show. Using the same RAM for all tests eliminates a variable, and all we do is apply the built in XMP profile. We believe this replicates the sort of system the vast majority our enthusiast audience would be running.
     
    edzieba and MLyons like this.
  11. MLyons

    MLyons 70% Dev, 30% Doge. DevDoge Staff Administrator Super Moderator Moderator

    Joined:
    3 Mar 2017
    Posts:
    3,637
    Likes Received:
    1,781
    TLDR: Dogbert is a shill
     
Tags: Add Tags

Share This Page