1. OCAU Merchandise now available! Check out our 20th Anniversary Mugs, Classic Logo Shirts and much more! Discussion here.
    Dismiss Notice

Ryzen 4000 Series Discussion

Discussion in 'AMD x86 CPUs and chipsets' started by Skramit, Nov 7, 2019.

  1. Elyzion

    Elyzion Member

    Joined:
    Oct 27, 2004
    Messages:
    7,374
    Location:
    Singapore
    Just a reminder, Zen 1 was 14nm, Zen 2 is 7nm. So it's not flat out impossible that AMD can't jump to 5nm for Zen 3.

    This is TSMC's investment, not AMD's.
     
  2. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    6,833
    Location:
    Dark City
    Ryzen 4000 is what 18 weeks away.

    There is no way to tape out an entire new process in 18 weeks, sample and get it out to market. Not enough time. They would barely have enough time to release a bug fixed cpu in time. Ryzen 4000 is using an improved 7nm process, which TMSC specifically designed to be easy to adopt.

    That doesn't mean there may not be 5nm Ryzen 4000 CPU's in the future. If AM4 has a big enough following (and it looks like it will) and DDR5 is slow to grow, I could certainly see AM4 and AM5 versions of the same die. Remembering AMD has moved all the Socket system IO off die.

    AM4 is quite likely to be a competitive low end socket for a while yet. Its conceivable that Ryzen 5000 and maybe 6000 APU's will be a socket AM4 thing. The whole concept of APU is to save money, real estate etc. DDR4 will likely help with that.
     
  3. Elyzion

    Elyzion Member

    Joined:
    Oct 27, 2004
    Messages:
    7,374
    Location:
    Singapore
    AMD is already got zen 5 in development... so what’s to say they didn’t work with TSMC for zen 3 on 5nm for the past 2-3 years? At the end of the day we have no clue until they officially announce it.
     
  4. RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    15,705
    Location:
    Brisbane
    The rumour about Zen 3 on 5nm was associated about TSMC stepping away from work with Huawei. So not in any long term plans for AMD. I doubt that AMD could take advantage on a short term basis.
     
  5. Elyzion

    Elyzion Member

    Joined:
    Oct 27, 2004
    Messages:
    7,374
    Location:
    Singapore
    At the end of the day, who cares, 4000 gonna be awesomeeee

    3a9843fd07bc27da383757ecf32716cd60ab7d12aad321859d174e5b9146256f.jpg
     
  6. TRG.dOinK

    TRG.dOinK Member

    Joined:
    Jun 27, 2002
    Messages:
    4,540
    Location:
    Melb East
    Last edited: May 29, 2020
  7. adamsleath

    adamsleath Member

    Joined:
    Oct 17, 2006
    Messages:
    18,801
    Location:
    Sunnybank Q 4109
    Nethiuz likes this.
  8. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    6,833
    Location:
    Dark City
    We are coming to the end of die shrinks.. I wouldn't be surprised if we hit it with AM5.

    5nm > 3nm > ~1nm.. 0.5nm would be 14 atoms per transistor, so you are getting pretty close to the hard limit, and the soft limit will be well before that. And shrinking has inevitably had side effects, not the continued benefit we expect.

    That being said. 5nm seems to be highly desirable. 80% greater density. AMD can make a much larger/complicated chip for less than intel. And we have barely exploited parallelism. AMD is still seeing significant IPC improvements with each generation. We are also seeing significant power per watt improvements
     
    SnooP-WiggleS likes this.
  9. MUTMAN

    MUTMAN Member

    Joined:
    Jun 27, 2001
    Messages:
    7,966
    Location:
    4109
    thats a 2D dimension of a 3D thing.
    i doubt we're close to the end just yet ( while acknowledging it is there somewhere on the horizon )
     
  10. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    6,833
    Location:
    Dark City
    https://www.princeton.edu/~maelabs/mae324/glos324/silicon.htm

    So we are getting pretty close to the hard end of conventional silicon. You aren't going to get anything smaller than 0.25 nm because then that is smaller than an atom to atom spacing, they can do many tricks but at some point you run out of spacing.

    So 1-2nm is probably the realistic limit in bulk making billions of transistors out of silicon. You might be able to squeeze that very slowly over the years with some tricks, but certainly don't bank on there being a hundred more die shrinks.

    https://arstechnica.com/gadgets/2016/07/itrs-roadmap-2021-moores-law/

    You can stack more layers, but they already do that and have done that for decades. There is a limit to the stacking thermally, electrically etc. Stacking layers makes much more sense for memory than it does for CPU processing.

    https://www.electronicdesign.com/te...d-qlc-push-flash-memory-technology-boundaries

    Which is why I am saying things like its entirely possible AM5 will be the last AMD socket to get die shrinks. Its not the end of computing, you can make much bigger CPU's, better designed CPU's, quantum technologies etc.
     
  11. RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    15,705
    Location:
    Brisbane
  12. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    6,833
    Location:
    Dark City
    Fab costs are also spirally exponentially.

    I think its something like $30 billion per fab now.
    Which is why there are only 3 in the world doing it. Likely to only be 2 successful at <5nm.

    It gets exponentially harder and more expensive for each shrink now. AMD dual die strategy makes a lot of sense

    Which is why suddenly getting a 5nm design out the door in a few months to take advantage of a window is impossible. 5nm may end up being the commercial limit for regular processor designs and further shrinks may only be feasible for applications like memory.
     
  13. dirtyd

    dirtyd Member

    Joined:
    Jan 4, 2006
    Messages:
    4,053
    Location:
    Melbs
    The x nm figure is just that, a figure. Nothing in a 7nm process is actually spaced at 7nm, these days it's just a relative measure to previous nodes.

    Considering how far things progressed with DUV, and how early the industry is with EUV adoption and maturation, seems like a fair bit of learning and optimisation to come purely from the light source.
     
    museumman likes this.
  14. cbwolf

    cbwolf Member

    Joined:
    Dec 23, 2001
    Messages:
    2,737
    Location:
    Melbourne
    I was looking at laptops today, and holy bejesus the Ryzen 4000 mobile CPUs just completely shit all over anything that Intel has out currently. It's actually amazing.

    I didn't realise how far ahead AMD were technologically with their CPUs honestly. I'm surprised. Intel seems like they really fell asleep and let AMD sneak up and fist them.
     
    nCrypt and OJR like this.
  15. theSeekerr

    theSeekerr Member

    Joined:
    Jan 19, 2010
    Messages:
    3,461
    Location:
    Broadview SA
    Part of it was complacence, but a lot of it was just Intel getting really, really unlucky with the development of their 10nm process. They still haven't got it working right...at the moment they're actually expecting they'll have mainstream chips out on the next process the same year they finally manage to do that on this one.
     
  16. jpw007

    jpw007 Member

    Joined:
    Nov 23, 2008
    Messages:
    4,642
    Location:
    Melbourne
    Can/will we see the end of silicon and something else replacing it that for the same density is much better, so we go back to something that's 45nm but it's as powerful and efficient or whatever as the 5nm silicon...then we start the process of shrinking those dies down again?

    (I have nfi abou any of it and i need to sleep, but hey you guys are intelligent and can tell me if i'm an idiot or not but it sounds cool in my sleep deprived brain!)
     
  17. asho444

    asho444 Member

    Joined:
    Feb 7, 2012
    Messages:
    1,713
    Location:
    Sunny Coast
    Wasn't graphene going to be the next big thing but just much more expensive?
     
  18. Sledge

    Sledge Member

    Joined:
    Aug 22, 2002
    Messages:
    8,318
    Location:
    Adelaide
    it always is when it's new...
     
  19. Maldark

    Maldark Member

    Joined:
    Jul 19, 2004
    Messages:
    642
    Location:
    Newcastle, NSW
    Photonic computing will be next. Replace the electrical signals with light ones, now you're no longer bound by the momentum of electrons.

    It's still way early days though, the first transistors are only now being modelled and tested.

    Even if we switched to another material and kept with electronic computing, there's physical limits in place we simply can't bypass.
     
    jpw007 likes this.
  20. RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    15,705
    Location:
    Brisbane
    Quantum computing is slowly gathering steam, but not sure how it will work, if at all, for general computing.
     

Share This Page

Advertisement: