Very Excited About Upcoming 7nm Ryzen, but...

Discussion in 'AMD Processors' started by Zarathustra[H], May 6, 2019.

  1. tangoseal

    tangoseal [H]ardness Supreme

    Messages:
    7,006
    Joined:
    Dec 18, 2010
    Zen 2 12 core is going to smoke 9900k
     
  2. defaultluser

    defaultluser [H]ardForum Junkie

    Messages:
    12,186
    Joined:
    Jan 14, 2006
    Yeah, it sounds like Zen 2 8 core is going to match the 9900k and embarrass the 9700k, so prices on that will have to fall significantly. I expect Zen 2 8-core to launch at the same price as 2700x is currenty, leaving plenty of room up-top for the more expensive 12-core, and possibly 16-core on launch day.
     
    Last edited: May 15, 2019
  3. Boil

    Boil [H]ard|Gawd

    Messages:
    1,273
    Joined:
    Sep 19, 2015
    AMD said the AM4 socket was good to 2020...

    TO 2020, not thru 2020...

    So maybe sometime in 2020 we get an AM4+ / AM5 socket, along with quad-channel DDR5 memory (and maybe high-speed low-latency 32GB & 64GB sticks)...?!?
     
  4. defaultluser

    defaultluser [H]ardForum Junkie

    Messages:
    12,186
    Joined:
    Jan 14, 2006
    Nope, AM5 wil likely l remain dual-channel DDR5. The secret to this is that DDR5 is using a dual bus for better utilization of each memory module, giving you about 35% actual bandwidth at the same bus transfer rates. And it's looking to START at 3200, and jump all the way up to 6400.

    Unfortunately, they wil need a new socket for the new pin outs for DDR5, and will likely pair it with 5nm EUV Zen 3 in 2021/2022. Zen2+ will still be on DDR4, and will probably not touch much.

    Right now DDR5 DIMMs will be available by the end of the year, and they are expected to show up in servers first, then high-end cell phones and mainstream desktops. AMD's independent I/O die means that you can use DDR5 on EPYC server refresh starting next year, while still using the same refresh chips on DDR4 mainstream systems.
     
    Last edited: May 16, 2019
    RAutrey, drescherjm and Pieter3dnow like this.
  5. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,516
    Joined:
    Jul 29, 2009
    https://www.tomshardware.com/news/what-we-know-ddr5-ram,39079.html

    Sorry for the THG link :) . But it explains the numbers a bit better :) .
     
    defaultluser likes this.
  6. Zarathustra[H]

    Zarathustra[H] Official Forum Curmudgeon

    Messages:
    27,667
    Joined:
    Oct 29, 2000
    In highly threaded encoding/rendering workloads, certainly.

    The question is what happens in more typical workloads, when you are more likely to pin a single core, with much lower utilization on other cores.

    Single core Cinebench scores are what I am going to be looking for as a better predictor of general purpose performance.
    I'm excited to see where 7nm Ryzen lands in this regard.
     
  7. Rockenrooster

    Rockenrooster Limp Gawd

    Messages:
    187
    Joined:
    Apr 11, 2017

    Single core benchmark as a better predictor of general performance???

    You need to look at both of them to make a good decision.

    I'd say they're equally important or else we'd be thinking unlocked dual cores are the way to go because their higher single threaded performance was a better indicator of general perf!

    The people here looking to buy the higher core count CPUs are not going to be paying attention to single threaded perf AS much as MT. However I'm 100% certain that AMD will have a higher clocked CPU with less cores for the people that like ST to be high.
     
    defaultluser likes this.
  8. VanGoghComplex

    VanGoghComplex [H]ard|Gawd

    Messages:
    1,658
    Joined:
    Apr 5, 2016
    Honestly, if the 12c 24t processor lags just a little behind in single threaded performance, I will still jump all over it. 300% core count increase over my 7700k... Yes please!
     
  9. Zarathustra[H]

    Zarathustra[H] Official Forum Curmudgeon

    Messages:
    27,667
    Joined:
    Oct 29, 2000

    Not what I intended to say.


    My point is that you know how many cores you have. You don't need to test to verify this. Scaling with added cores isn't all that different from arcitecture to architecture.

    If you need more cores, choose the model with more cores, if you need fewer cores, choose the model with fewer cores.

    Where the key difference comes in is how capable each core is. "IPC"*clock speed.

    Very rarely have I had a workload that is limited by my number of cores. Typically when I am CPU limited (which is rare) it is because I have a bunch of cores that are either idle or loaded no more than ~50% and one core that is pinned at 100% holding everything back.

    This situation is improved by adding more per core performance, NOT by adding additional cores.
     
    Last edited: May 16, 2019
  10. Rockenrooster

    Rockenrooster Limp Gawd

    Messages:
    187
    Joined:
    Apr 11, 2017
    That makes more sense.
     
  11. Zarathustra[H]

    Zarathustra[H] Official Forum Curmudgeon

    Messages:
    27,667
    Joined:
    Oct 29, 2000
    Apart from my many typos :p