Ryzen 6000/Warhol Speculation

Discussion in 'AMD x86 CPUs and chipsets' started by RnR, Apr 23, 2020.

  1. SnooP-WiggleS

    SnooP-WiggleS Member

    Joined:
    Sep 20, 2004
    Messages:
    3,162
    Location:
    Ballarat, Victoria
    So it'll just be a zen3+ minor refresh, AM4 lives a little longer!
     
    MUTMAN likes this.
  2. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    7,371
    Location:
    Dark City
    Am4 for ever!!
    B550 seems pretty nice in terms of longevity if it goes another cycle.
     
  3. mesaoz

    mesaoz Member

    Joined:
    Jan 15, 2015
    Messages:
    7,376
    Location:
    South East QLD
    Hmm... so this year will be 7nm+ and next year will be 7nm++? Then 5nm 2022

    I don't like seeing that ++, reminds me of intel's 14nm+++++++++++++++
     
  4. OP
    OP
    RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    16,399
    Location:
    Brisbane
    At least we can see the end of the '+'s. N5 is a proven process, otherwise Apple would be spitting... umm... chips...

    And TSMC will have 2 more +'s for N5... N5p and N4.
     
    mesaoz likes this.
  5. Comrade VPZ

    Comrade VPZ Member

    Joined:
    Sep 18, 2017
    Messages:
    76
    Location:
    Sydney
    Yeah this sort of thing gives Intel time to catch up.
     
  6. mesaoz

    mesaoz Member

    Joined:
    Jan 15, 2015
    Messages:
    7,376
    Location:
    South East QLD
    I'm all for that. Intel needs to get their shit in order so they both have a reason to keep making better chips... Before the ARM overlords push them out.
     
    Comrade VPZ likes this.
  7. lithos

    lithos Member

    Joined:
    Oct 2, 2010
    Messages:
    6,532
    So, is Intel gonna call their 10nm+++++++++++++++++++++++++ chips "Solanas"?
     
    MUTMAN likes this.
  8. chook

    chook Member

    Joined:
    Apr 9, 2002
    Messages:
    3,789
    Pluses == performance.
     
  9. OP
    OP
    RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    16,399
    Location:
    Brisbane
    Warning.. brain farts and tired Sunday ramblings. All of the following have been pulled directly from my rear :)

    So over in twitter land a while back, some Korean/Chinese leaker pondered... "Why Warhol?"

    upload_2020-10-11_19-27-51.png

    Its looking pretty boring there. A Zen 3+ design. No N5. Even N6 seems to be ruled out since Rembrandt has it, but Warhol doesn't. No DDR5, nor PCIe5.

    Van Gogh, Rembrandt and Raphael all seem to have some nice inclusions. CVML (computer vision, machine learning) bits, DDR5... even Raphael seems to have Navi bits if that little green square can be trusted!

    So what could AMD pack into this 'boring' release? Lets check out the AMD release cadence for the Zen family and see how that slots in to the above chart...

    Code:
    Summit Ridge    Zen 1   Feb 2017
    Pinnacle Ridge  Zen+    Apr 2018    +14 months
    Matisse         Zen 2   Jul 2019    +15 months
    Vermeer         Zen 3   Nov 2020    +15 months
    Warhol          Zen 3+  Feb 2022    +15 months*
    Raphael         Zen 4   Jun 2023    +15 months*
    
    * - expected
    Seems too long to be honest for Warhol. I would expect AMD to be releasing N5 parts in 2022, and thats when Raphael is shown above to be released in. Zen 4 has been confirmed to be using N5 by AMD.

    upload_2020-10-11_19-52-43.png


    So why Warhol? Maybe because marketing needs a yearly release? Could they do a quick spin to have a release in 2021? And what could you do as a quick spin?

    1) IPC. But not much. No new instruction sets.

    2) 'Dark silicon'. Lets assume that AMD will get to use N6 for this release despite what it says in the first chart above. Thats ~20% extra density available. Thats more than in the step between GF14 to GF12 for Zen->Zen+. And back then AMD also used dark silicon to push the old design harder and faster.

    upload_2020-10-11_20-1-55.png
    Taken from https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/2

    Same design, same die size, just features scaled smaller with more (thermal/electrical) space between performance sensitive features, allowing them to clock higher. N6 is IP compatible with their previous releases, so no great workload.

    3) Better Infinity Fabric. Buildzoid touched on this in his recent ramble on the 5000 series release. The IO die has a skinny bus clocked quite high. AMD could widen the bus, providing more bandwidth, but they could also spec a better substrate to get higher clocks. A higher clocked IF makes some sense given that in new motherboards about to be released they are advertising memory support of 5800-6000Mhz. AMD still has a contract with GlobalFoundries (out to 2024) and all the IO dies have been built using their 12LP process. Last year GF announced 12LP+ allowing +20% speeds at the same power profile, and it will be available in volume in 2021. Just perfect for Warhol and a faster IO die. Its even IP compatible!

    And thats it. A nice farewell chip for the long lived AM4 socket, slotted into mid-2021 to make the marketers happy, but perhaps also an overclockers delight. Leaving 2022 for the next marketers golden trifecta of AM5, DDR5, TSMC N5 and perhaps even PCIe5 :D

    Interested in your thoughts. What else could AMD do without touching too many things? Or have I missed something bleeding obvious? :)
     
    Last edited: Oct 11, 2020
    adamsleath, MUTMAN and SnooP-WiggleS like this.
  10. chook

    chook Member

    Joined:
    Apr 9, 2002
    Messages:
    3,789
    Completely non-technical but I foresee a lot of mess in the marketing offices around releasing Zen 5 (I would call the Zen 3 refresh Zen 4 just so I could do this) in socket AM5 with DDR5, PCIe 5.0 on the N5 process and having 5 May be the release day. Marketing would already be working themselves into a lather over this possibility.
     
    MUTMAN likes this.
  11. OP
    OP
    RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    16,399
    Location:
    Brisbane
    That horse has bolted :)

    Another horse that I think has bolted is that Warhol will get official 5Ghz support. No worries on that score. Its hard to get all your ducks lined up :D
     
  12. adamsleath

    adamsleath Member

    Joined:
    Oct 17, 2006
    Messages:
    18,946
    Location:
    Sunnybank Q 4109
    must say. this is rapid development from amd, considering the 14nm intel chugalug. rumours than 12th gen intel will be 2H 2021, on their 10nm. with big/little cores.



    5nm in 2022? wtf?

    smartphones are pretty advanced too arent they.
     
    Last edited: Oct 12, 2020
    MUTMAN likes this.
  13. nope

    nope Member

    Joined:
    May 8, 2012
    Messages:
    3,314
    Location:
    nsw meth coast
    Andy Whorhol?
     
  14. OP
    OP
    RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    16,399
    Location:
    Brisbane
    Yeah AMD is using artist names for their non-server codenames.
     
    nope likes this.
  15. MUTMAN

    MUTMAN Member

    Joined:
    Jun 27, 2001
    Messages:
    8,576
    Location:
    4109
    so i've been considering your thoughts on this release for a while. just throwing my thoughts out here. and basically concur with what you've outlined.

    it'll be a small tweak to the chiplets. amd will take the opportunity to tweak and fiddle, no doubts about it.
    but the big change will be to the IO die. a die shrink would;
    1. efficiency, they can grab a few % on perf/watt.
    2. getting everything around the IF and memory controller ready for the higher clocks of ddr5
    3. another set of heels for Lisa, chook will get the old ones :lol:
     
    RnR likes this.
  16. OP
    OP
    RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    16,399
    Location:
    Brisbane
    Yeah a die shrink of the IO die with a 40% power saving is probably what AMD would be targeting. Extra watts that can be put towards the cpu die itself.

    I wrote my thoughts above before Vermeer and the associated news came out about how AMD was still on standard TSMC N7 process. The picture above then makes sense. Warhol, Zen 3+, will probably be on TSMC's N7p which will allow AMD to get an additional 7% higher frequencies at the same power profile. With a few extra watts stolen from the IO die power budget, we might be getting 6950X's hitting 5.5Ghz.

    TSMC's N6 won't be used for Warhol, as this process needs EUV machines, and those are in extremely high demand (ie N5 process), so no dark silicon tactics needed. N7p is still DUV based, so would be quite cheap comparatively, if it wasn't for the sheer demand.

    Another piece of evidence for the thought that it will be a 'small-tweak-release' is that Warhol doesn't appear at all on the serverside cpu lineup. Optimisations made for Zen 4 on the TSMC's N5 process will not be backported to Zen 3 and N7p. This would require multiple validation work for the same feature sets which would be a waste of money.

    Still no clue as to whether it will be released under AM4 or AM5. Nor adoption of DDR5. Too murky for me :)
     
    MUTMAN likes this.
  17. MUTMAN

    MUTMAN Member

    Joined:
    Jun 27, 2001
    Messages:
    8,576
    Location:
    4109
    i'll roll the dice ...
    AM4/DDR4 only
    a lovely farewell to the awesome comeback socket :thumbup:
     
  18. OP
    OP
    RnR

    RnR Member

    Joined:
    Oct 9, 2002
    Messages:
    16,399
    Location:
    Brisbane
  19. MUTMAN

    MUTMAN Member

    Joined:
    Jun 27, 2001
    Messages:
    8,576
    Location:
    4109
    I'm loving the idea of a 8 core zen with navi gpu APU being fed by 8400 memories....

    AMD shouldn't/wouldn't miss this opportunity ? Shirley ?
     
  20. MUTMAN

    MUTMAN Member

    Joined:
    Jun 27, 2001
    Messages:
    8,576
    Location:
    4109
    but this means warhol is ????
     

Share This Page

Advertisement: