Page 670 of 876 FirstFirst ... 170570620660668669670671672680720770 ... LastLast
Results 6,691 to 6,700 of 8752

Thread: Official AMD Zen uarchitecture thread

  1. #6691
    Banned
    Join Date
    Oct 2010
    Posts
    490
    Quote Originally Posted by french toast View Post
    Zen 2 2018 H1 on 14nm lpp
    zen 3 2019 H1 on 7nm.
    yeah this is the most realistic and likely scenario. Zen+ / Zen 2 in H1 2018 on 14nm and Zen++ / Zen 3 on GF 7nm in H1 2019.

  2. #6692
    Senior Member
    Join Date
    Jan 2013
    Posts
    706
    Quote Originally Posted by inf64 View Post
    One cannot go wrong with traditional 10-15% . It depends whether they go for full 256bit L/S units which would double the cache BW and make FP unit 2x256bit capable. I'm not sure whether they can do that on 14nm and keep the clocks this high (reaching 4Ghz).

    Maybe they can just get a ~10% IPC doing PD/XV type of changes and bump the CCX to 6 cores. That would be enough to stay competitive in 2018 until they get 7nm ready.
    ALU/AGU parity

  3. #6693
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,487
    I expect Zen 2 design is pretty much complete by now. This will be able to take advantage of the low hanging fruit for improvements, they've been prodding Zen for over a year now to see where the issues are.

    My guess is:

    * Launch in Q1 2018
    * 14nm LPP
    * 8C - 16T probably, but SMT4 could be a pleasant surprise for server SKUs
    * 10% IPC boost, but little clock boost (5%?) - but AMD will squeeze Intel's IPC lead
    * +10% die size (or CCX size)
    * 4-core CCX

    Q4 2018 shrink to 7nm, with a 12C SKU - initially for servers (48-core Starship) - 36-48MB L3.
    Maybe this is where SMT4 will be introduced (server only).
    Consumer 7nm SKUs during 2019, Q1 would fit in nicely for an annual Zen event.
    If these are Zen 3 I expect maybe 5% IPC boost (15-20% normalised Multithreaded Throughput boost if SMT4).
    It may make sense to move to a 6-core CCX, but OTOH maybe 3 4-core CCXs will provide the same effect with a faster fabric.

    Maybe time for a new thread for Zen 2?

  4. #6694
    640k who needs more?
    Join Date
    Mar 2010
    Posts
    995
    Quote Originally Posted by schroeder View Post
    ALU/AGU parity
    I think the ALU/AGU ratio will stay the same of course, 4ALUs and 2 AGUs.

  5. #6695
    Senior Member
    Join Date
    Jan 2012
    Posts
    829
    Quote Originally Posted by sdlvx View Post
    I think they would be keeping quiet. They have no reason to release Vega now, as it'd just encourage people to build computers with Vega and Intel instead of Vega and Ryzen. Not to mention look how quiet AMD has been with Ryzen.

    AMD can coast on Zen for a year or two. I'd really like to see another AMD upset, when they move to 7nm and every price bracket gets cores added. i.e. SR3 1100 4c/4t @ $129 turns into SR3 2100 6c/6t @ $129; 8c/16t moves to 12c/24t at same price brackets, etc.

    AMD could bleed Intel out if they followed a plan like this. It would hurt AMD's potential profits, but anything would be better than how their CPUs have been for the last few years.

    Neutering Intel by attacking their large margins and reducing their R&D would give AMD a huge advantage. AMD has the potential to turn Intel's private fabs into a memory. Intel needs a lot more R&D money. They have to maintain those fabs and keep shrinking, and no doubt they're spending CPU revenue on other pet projects that all fail.

    I see good reason for AMD to push to 7nm sometime in 2018. AMD doesn't need the R&D money Intel needs, and they can capitalize on it. They're clearly getting aggressive with Intel, just imagine what these low prices will do to Intel's contra-revenue programs. When Athlons were expensive, Intel could afford to slide some money under the table and coerce OEMs to switch to Intel. But it's going to take a lot more to coerce OEMs to skip on AMD with the prices being this low.

    It would be in AMD's best interests to get more cores on 7nm out by the end of 2018. Then again, I'm hoping for a core war. But I suspect Ryzen is going to be the end of Intel gouging on mainstream quads.

    EDIT:


    Oh no why did you post this our resident marketing slide technology expert will show up to measure pixels and give us Zen+ IPC improvements.
    I dont believe they would to take the risk, 14m is decent performance and is nice and safe for a couple of years, why take the risk of another 20nm?

  6. #6696
    8-bit overflow
    Join Date
    Oct 2010
    Posts
    272
    I think you guys are putting to much faith in 7nm ramp. I expect a PD like refresh ~12months from now, so 5-10% ipc and a few 100more mhz on clock still on 14nm LLP. The next one ( 12months later) will be more interesting, i think it will still be 14nm, could it be LLP, or maybe HP? 6core CCX? 256bit L/S/SIMD? AVX-521 support (over 256bit units, mainly for scatter gather)? i dont know.

    Then i think the 3rd generation which is another 12months away will be on 7nm.

  7. #6697
    Senior Member
    Join Date
    Jun 2010
    Posts
    4,228
    Quote Originally Posted by GeorgiD View Post
    DCO, yes, AMD have to release more products later this year - Vega, Raven Ridge, Naples... but what about next year? Nothing?
    Yields can improve without releasing new faster versions - they can instead lower the power consumption on the already released chips...
    From cpus I would expect changes like from Trinity to Richland, minor ones.
    I think most of the next year is going to be APUs, they will be probably ready by the end of the year or the beginning of the next year, Zen is getting ready to be released now, Vega is still in the oven and the APU brings both technologies.
    GPUs I can-t say, I am not sure what they are going to release this year and at what prices.

  8. #6698
    Senior Member
    Join Date
    Nov 2011
    Posts
    2,593
    Quote Originally Posted by french toast View Post
    I dont believe they would to take the risk, 14m is decent performance and is nice and safe for a couple of years, why take the risk of another 20nm?
    Because whenever AMD gets ahead, or gets close, Intel strikes back and AMD is back to where they started. They need to be aggressive and keep Intel on the defensive if they want to have long term success that lasts more than a year or two.

    AMD can easily bleed Intel out. Intel spends a lot more to get a CPU released.

    Not only that, I don't think these Ryzen prices are anything special. They are where the market should be if Intel didn't spend the last few years doing absolutely nothing because they thought they were fine. Ryzen prices are sane. People are thinking the Ryzen prices are great and a massive value. If Intel was not abusing customers for the last 5 years, we would already be at Ryzen prices for performance.

    I would even use Intel's current situation as a reason why you shouldn't sit on your laurels for too long, it makes it easier for everyone else to catch up. I have a good feeling Intel's arrogance when competing against construction cores will go down in history as a massive blunder. And there's no reason for AMD to get arrogant or lazy because they are finally competitive. Intel will answer back, it may take two or three years, but it's coming. And there's no point in AMD sitting around waiting for it to happen, because Intel usually wins when the second battle comes in a war like this.
    Within 27% margin of error and 3mm off my estimate of something 4mm, good enough to predict the future

  9. #6699
    Senior Member
    Join Date
    Jan 2012
    Posts
    829
    Quote Originally Posted by sdlvx View Post
    Because whenever AMD gets ahead, or gets close, Intel strikes back and AMD is back to where they started. They need to be aggressive and keep Intel on the defensive if they want to have long term success that lasts more than a year or two.

    AMD can easily bleed Intel out. Intel spends a lot more to get a CPU released.

    Not only that, I don't think these Ryzen prices are anything special. They are where the market should be if Intel didn't spend the last few years doing absolutely nothing because they thought they were fine. Ryzen prices are sane. People are thinking the Ryzen prices are great and a massive value. If Intel was not abusing customers for the last 5 years, we would already be at Ryzen prices for performance.

    I would even use Intel's current situation as a reason why you shouldn't sit on your laurels for too long, it makes it easier for everyone else to catch up. I have a good feeling Intel's arrogance when competing against construction cores will go down in history as a massive blunder. And there's no reason for AMD to get arrogant or lazy because they are finally competitive. Intel will answer back, it may take two or three years, but it's coming. And there's no point in AMD sitting around waiting for it to happen, because Intel usually wins when the second battle comes in a war like this.
    im not saying they should sit on their laurels but due to their financial situation these past years they cant have another failed expensive node, i believe they will only move when the risk is mittigated somewhat.

    2019 H1 is absolutely fine and would be parity with intel.

  10. #6700
    Senior Member
    Join Date
    Jan 2010
    Location
    London
    Posts
    3,213
    Shouldn't SMT4 require much deeper redesign/widen up the core to keep them threads fed? I'm not a CPU designer, but don't think you can take Zen core and just slap 2 more threads per core just by doing minor changes and improvements.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux