Become a Patron!

Intel Says AMD EPYC Processors "Glued-together" in Official Slide Deck

Discussion in 'AMD Processors' started by JustReason, Jul 12, 2017.

  1. juanrga

    juanrga Gawd

    Messages:
    551
    Joined:
    Feb 22, 2017
    Wait because this can be more hilarious still. This emphasis from certain media to focus on the "glued-together" wording found in one slide of Intel presentation looks like a huge FUD campaign to divert attention from the important stuff: that Skylake Xeon is very fast, efficient, selling well. and customers are happy with it.

    The same media is ignoring the rest of slides of the presentation including the next one with customers sharing their own benchmarks on SKL Xeon

    [​IMG]
     
    Last edited: Jul 16, 2017
  2. aztekk

    aztekk Limp Gawd

    Messages:
    148
    Joined:
    Nov 4, 2014
    I really hope not, soldering (more PC than gluing?) dies onto a flip-chip package is hardly the most efficient way to increase core count. If you do this to combat the thermals of a die then eventually you will run into a very hot package with potential to compromise solder balls akin to Tesla 1.0, perhaps why Intel abandoned it in those days. I'm sure both AMD and Intel are going to develop more adequate ways to move forward after Moore's Law.

    I don't think AMD is going to win much of the server market. I hope I am wrong and we see cheaper dedicated servers but with CPU's being a miniscule part of overall cost there (and very conservative customers) I don't see it happening.

    Regarding "glue" we could argue about the chemical properties of TIM being similar to certain glues used in electronics, but I won't go there. Intel also did what AMD is doing with Pentium D and Core2Quad, so it's hypocritical of them to be focusing on that point alone. Shittalking your competitor in general is a bad marketing move, and since this is not a consumer product launch, I question the necessity of this kind of advertising.

    However it seems to have been successful, since everyone is now reporting these slides that would otherwise not been shown outside of conference rooms providing more publicity for Skylake-SP. Intel was able to capitalize on AMD fanboys blowing a gasket and thus captured the news cycle which would otherwise be saturated by AMD Epyc and Threadripper. Bravo! :)

    With that said I think there is too much time allocated to arguments amongst the enthusiast/gaming community regarding products that were never meant for that market segment.
     
    spine and juanrga like this.
  3. sirmonkey1985

    sirmonkey1985 [H]ard|DCer of the Month - July 2010

    Messages:
    19,214
    Joined:
    Sep 13, 2008
    well we got nothing else to do with our time until trustworthy threadripper benchmarks come out so we might as well debate about this crap, lol.
     
  4. JustReason

    JustReason [H]ard|Gawd

    Messages:
    1,904
    Joined:
    Oct 31, 2015
    Still not proof, this is a marketing slide. And the articles only spoke to the wording of the title. most only linked the slide deck with the forum posters posting the slides. Doesn't change the rest of the slide deck where Intel is shaking in their boots and tries with all their might to smear EPYCs image. Whether you wish to acknowledge it or not, AMDs outlook amongst the community including IT professionals from all markets is GOOD.
     
    Last edited: Jul 16, 2017
  5. n=1

    n=1 2[H]4U

    Messages:
    2,276
    Joined:
    Sep 2, 2014
    1. It's only FUD if it's based on false/inaccurate information. That marketing slide deck isn't a work of fiction, so the use of the word "FUD" is completely inappropriate.

    2. If you have a product that is objectively superior in every single metric, then simply show the benchmarks and don't sling mud at your competitor.
     
    KazeoHin, ManofGod and JustReason like this.
  6. chenw

    chenw 2[H]4U

    Messages:
    3,529
    Joined:
    Oct 26, 2014
    Ironically, I'd be a LOT more impressed if EPYC was really glued together.

    Just think what AMD could have done if they used duct tape instead...
     
    juanrga, Chimpee and KazeoHin like this.
  7. gigaxtreme1

    gigaxtreme1 2[H]4U

    Messages:
    2,743
    Joined:
    Oct 1, 2002
    PCS are manufactured with a kind of therosetting plastic like epoxy or a vinyl ester. Glue is the wrong word and Intel looks bad for dissing AMD.
     
  8. KazeoHin

    KazeoHin [H]ardness Supreme

    Messages:
    6,354
    Joined:
    Sep 7, 2011
    Intel is shaking in their boots. They've forgotten how to compete.
     
  9. n=1

    n=1 2[H]4U

    Messages:
    2,276
    Joined:
    Sep 2, 2014
    ^that's going a bit too far.

    I'd say the mud slinging from Intel simply means they recognize EPYC as a very tangible threat to their Xeons. I mean we never heard a peep out of Intel back in the Bulldozer days did we?
     
  10. Shintai

    Shintai [H]ardness Supreme

    Messages:
    4,349
    Joined:
    Jul 1, 2016
    They did both with Barcelona and Bulldozer.
     
    juanrga likes this.
  11. juanrga

    juanrga Gawd

    Messages:
    551
    Joined:
    Feb 22, 2017
    1. The FUD part is when some people takes that slide and claims that EPYC is not getting desing wins because Intel is playing dirty.

    2. I have a better idea, leave potential customers to evaluate your product and get heir own conclusions. Intel did that, and customers are now sharing their experiences/benchmarks with SKL, whereas all what AMD has are slides and demos where the performance of Xeons is 'debatable'...

    The use of the term "glue togheter" doesn't imply the use of ordinary glue.

    In above examples, physicts, prosecutors, and language experts aren't implying that quarks, shreds of evidence, and words are joined with ordinary glue. When Intel says that EPYC are four dies glued together, Intel doesn't imply ordinary glue has been used.
     
  12. n=1

    n=1 2[H]4U

    Messages:
    2,276
    Joined:
    Sep 2, 2014
    Haven't seen any of that in here so far.

    Intel did more than just that when they included the "glued together" slide. This much is fact and not up for debate.
     
    JustReason likes this.
  13. BrotherMichigan

    BrotherMichigan n00bie

    Messages:
    7
    Joined:
    Apr 8, 2016
    It has been so cute watching juanrga, Shintai, and other get their shorts all in a bunch over Ryzen/EPYC. I consider the launch of both to be a complete success on that basis alone, glued together CPUs or no!
     
  14. SighTurtle

    SighTurtle Gawd

    Messages:
    822
    Joined:
    Jul 29, 2016
    http://www.tomshardware.com/reviews/intel-amd-die-fabric-slides,5125.html

    Tom's Hardware has taken the time to, in their own words,

     
  15. noko

    noko 2[H]4U

    Messages:
    2,591
    Joined:
    Apr 14, 2010
    I would disagree, getting core count up by using multiple sockets and even more racks will consume more space and power then combining more cpu's on a socket. If bigger more complex cpu's are made they too can be placed on interposer which also allows space savings, faster communications between the cpus. This maybe will get very interesting once EPYC becomes available, at least we will get a good idea now viable AMD has made it. Hopefully not too much growing pains will come with it.
     
    juanrga likes this.
  16. juanrga

    juanrga Gawd

    Messages:
    551
    Joined:
    Feb 22, 2017
    Right, it would go against the benefits of integration.
     
  17. Shintai

    Shintai [H]ardness Supreme

    Messages:
    4,349
    Joined:
    Jul 1, 2016
    There really isn't any power saving doing it. Its all the physical space size. And using interposers got their own limitations and cant really be scaled to any meaningful way. You would need something without the size limitation for that like EMIB stitching or other solutions. And without these and the proper interconnect you sit back with a lot of penalties. A good example on how important the integration is can be seen in the consoles. Their 190 cycle penalties pretty much made the last cluster useless.
     
    Last edited: Jul 18, 2017
    aztekk likes this.
  18. JustReason

    JustReason [H]ard|Gawd

    Messages:
    1,904
    Joined:
    Oct 31, 2015
    You know what sucks about this post, it says almost nothing. A bunch of inference without any clarification.

    Power savings? I think when Intel eventually releases their HEDT 16c and you can compare it directly to TR then we can have a better well informed analysis.

    Physical space size? Not sure what inane point you were attempting here.

    Interposers limitations and scaling. BS Limitations yes but that is definitely offset by scaling, we see this with EPYC and its PRICE against the fleecing of Intels customers.

    And what can be seen with the consoles? The consoles in no way are related to ZEN in any form.

    Maybe instead of these quips and vague negative tones you might enlighten us more with greater detail. Honestly doesn't look like you have much or are intentionally being daft to obfuscate any positives on the AMD front. It is known that what AMD has done with ZEN is where virtually every chip designer is going, just AMD got there first.
     
  19. Shintai

    Shintai [H]ardness Supreme

    Messages:
    4,349
    Joined:
    Jul 1, 2016
    Even comparing a 16C to 16C is pointless due to the different uarchs. At best you would have to compare socket counts. But here is a hit, try check an EPYC server board vs a Xeon and tell me how many phases there are on each.

    Consoles is the same issue, then you can pretend its otherwise.

    Even ARM is going against the Zen setup. ARM got an 8 core mesh with A75.

    What is this price you talk about? Consumer line prices was unaffected by Ryzen. And Xeon line is unaffected by Epyc. AMD Q2 result is on the 25th, then you yourself that hype and reality is 2 separate things.
     
    aztekk likes this.
  20. JustReason

    JustReason [H]ard|Gawd

    Messages:
    1,904
    Joined:
    Oct 31, 2015
    16c to 16c is far from pointless. You make the point about power usage in response to the IF design of AMD with vague innuendos toward high power usage, where likely that Intel 16c is going to use far more power and clocks may very well be in question as well. Look at the 10c from Intel right now, adding 6 cores is going to make it far worse, but unlike most of you I am willing to wait and see how it pans out.

    Price? Seriously it isn't hard to see that Intel has been fleecing you for years yet you pretend they are some kind of godsend to computing. Amds prices for equivalent performance is far better and only goes so far to prove Intel has been raping their customers. Defending the very company that never let ethics or morality stand in the way of a meager dollar.

    And again these short quips are saying nothing, just a bunch of fear mongering inspired by ignorance and some unfounded desire to quell any positive views on a product proving to be far more superior than your expectations.
     
  21. n=1

    n=1 2[H]4U

    Messages:
    2,276
    Joined:
    Sep 2, 2014
    http://www.tomshardware.com/reviews/intel-amd-die-fabric-slides,5125-2.html

    Lordy someone please tell me this slide deck is a work of fiction :ROFLMAO:
     
    Kyle_Bennett and juanrga like this.
  22. SighTurtle

    SighTurtle Gawd

    Messages:
    822
    Joined:
    Jul 29, 2016
    juanrga likes this.
  23. n=1

    n=1 2[H]4U

    Messages:
    2,276
    Joined:
    Sep 2, 2014
    More proof the marketing and PR guys are complete idiots who don't have a clue what they're doing
     
    juanrga likes this.
  24. aztekk

    aztekk Limp Gawd

    Messages:
    148
    Joined:
    Nov 4, 2014
    +1 for EMIB, that would be what I was referring to, I don't know why he figured I was advocating adding more sockets. I stated adding more dies onto an interposer in the classic way is not scalable since 2005, and we need new technologies to overcome the complexity and performance issus with interconnect wires. EMIB is imo so far the best solution that is cost-effective:

    [​IMG]

    That said its a moot endeavor if your uarch's IPC is from the last decade, which is the more pressing reason why Epyc will fail.
     
    juanrga likes this.
  25. juanrga

    juanrga Gawd

    Messages:
    551
    Joined:
    Feb 22, 2017
    And both guys would be automatically fired for that.
     
  26. JustReason

    JustReason [H]ard|Gawd

    Messages:
    1,904
    Joined:
    Oct 31, 2015
    First you are in the wrong forum for Intel a**kissing. Second AMDs effective IPC is closer to Haswell , 4 yrs old. And one should never make baseless predictions as it usually comes back to bite you in your behind. EPYC will likely do quite well, not gonna break any records but with its price and I/O it will surely sell.
     
  27. Osjur

    Osjur Limp Gawd

    Messages:
    176
    Joined:
    Apr 26, 2012
    I don't get this... some peeps are saying EPYC is going to be huge success and will get big market share right away in data center market, other peeps saying they will not gain any market share at all because it will fail and there's nothing good about it. Don't you even remember (or maybe you are just too young) what happened when AMD actually released a SUPERIOR server cpu in the form of Opteron and how long it took to gain some meaningful market share. If my memory serves me, they went from single digit number at 2003 to little over 20% at 2006 when Core based Xeon was released. That's frigging four years with superior chip. And by then, they had over 50% market share in consumer market.

    And now we have a situation where they have a chip which is competitive but not superior and while we (hopefully) don't have the same kind of situation where money changes hands under the table like back in the days, it is still going the be a uphill battle against the blue giant, especially when the current mind share is still completely blue. I can say with quite big confidence that the place where I am working atm will not have a single AMD based server from any of our customers until next year. And that would still happen even with a superior chip as there are contracts, done-deals, rentals, you name it, in play which has been negotiated already. And lets face it, server customers are more conservative when it comes to products from "new" manufacturer.

    My prediction is that AMD might snatch about 5-8% server market share in the span of next 12 months (and I believe even that is a quite optimistic), and that's mostly from their 1S platform as it looks very appealing. That is the place where Xeon is at its weakest against EPYC
     
    juanrga and aztekk like this.
  28. SighTurtle

    SighTurtle Gawd

    Messages:
    822
    Joined:
    Jul 29, 2016
    There is a middle ground beyond huge success and utter failure, like Osjur says, 5-8% is possible, hell even 2-4% would be better than the 0.1% it has now, and it should bring much needed revenue to stabilize AMD for the future.
     
    n=1 and JustReason like this.
  29. Digital Viper-X-

    Digital Viper-X- [H]ardForum Junkie

    Messages:
    13,147
    Joined:
    Dec 9, 2000
    Isn't that what AMD said about Intel Quad cores when they launched the X4s :p?
     
  30. JustReason

    JustReason [H]ard|Gawd

    Messages:
    1,904
    Joined:
    Oct 31, 2015
    You might be mis-interpreting enthusiasm to mean huge marketshare when it mostly means no where to go but up. Hell 8% is a huge number. Server sales for most businesses isn't every year so it will take a few years for some of this to trickle into meaningful numbers. Likely most of the new Skylake Xeon sales were made months ago, not yesterday as some infer.
     
  31. Digital Viper-X-

    Digital Viper-X- [H]ardForum Junkie

    Messages:
    13,147
    Joined:
    Dec 9, 2000
    That's partially because intel was bribing / coercing OEMs to use their CPUs

    Additionally, people don't buy servers every year, or every 2 years. Usually we keep servers well past their service life as well, they just get demoted to non-critical path use. So buying new servers, it comes once every 5 years.
     
    Last edited: Jul 18, 2017
    JustReason likes this.
  32. gigaxtreme1

    gigaxtreme1 2[H]4U

    Messages:
    2,743
    Joined:
    Oct 1, 2002
    I believe the "coercion" was to NOT use AMD processors.
     
  33. Digital Viper-X-

    Digital Viper-X- [H]ardForum Junkie

    Messages:
    13,147
    Joined:
    Dec 9, 2000
    you know what I mean.
     
  34. JimmiG

    JimmiG 2[H]4U

    Messages:
    2,423
    Joined:
    Apr 3, 2008
    So how do you suggest they increase the core count in a post Moore's Law world? We'll probably get 7nm, maybe even 5nm, but after that, it's a dead end. Making the chip bigger and bigger will lead to lower yields and unmanageable thermals. Already, processors spend a huge amount of their power budget simply moving data around the chip to where it needs to be, and with a bigger chip, the power requirement would go up considerably.
    The other option would be to simply make smaller, less complex cores and put more of them on the chip aka Xeon Phi - great for some applications but clearly not a general purpose solution.

    From a power/thermals perspective, the MCM is indeed the most efficient way to increase core count. The Infinity Fabric only uses a couple of watts. Intel will switch to MCM's for their highest core count Xeons too in the next few years, I just don't see any other way for them to compete with Epyc core counts.
     
    noko likes this.
  35. juanrga

    juanrga Gawd

    Messages:
    551
    Joined:
    Feb 22, 2017
    It goes exactly in the opposite direction. The MCM approach introduces extra power consumption (compared to a monolithic die) due to propagating signals as

    die --> TSV --> interposser --> TSV --> die.

    That is one of the reasons why everyone else does monolithic server dies: Intel, IBM, APM, Sun/ORacle, Fujitsu, Cavium,, Broadcomm,...

    AMD uses MCM approach for reducing costs. They don't have the money to design and validate different dies for server/desktop.
     
    aztekk and Shintai like this.
  36. juanrga

    juanrga Gawd

    Messages:
    551
    Joined:
    Feb 22, 2017
    The hype around EPYC is disproportionate. I have seen idiots predicting that EPYC will get 15% server marketshare by Q4 this year. The company's own predictions are much lower. AMD expects to get around 10% of market for 2020 (including Zen2). And they are still being too optimistic. Fortunately some analysts are cutting down the hype:

    I have been saying something like the bold-green part for months.
     
    aztekk and Shintai like this.
  37. OrangeKhrush

    OrangeKhrush Gawd

    Messages:
    586
    Joined:
    Dec 15, 2016
    Getting a market footprint is the first step, any market share is a gain and important on.
     
  38. n=1

    n=1 2[H]4U

    Messages:
    2,276
    Joined:
    Sep 2, 2014
    lol analyst opinions

    If their predictions were worth a damn I'm sure they'd be out enjoying life rather than prognosticating the next big boom or bust ;)
     
  39. mnewxcv

    mnewxcv [H]ardness Supreme

    Messages:
    4,799
    Joined:
    Mar 4, 2007

    thats actually a screenshot from an unreleased NES game.
     
  40. noko

    noko 2[H]4U

    Messages:
    2,591
    Joined:
    Apr 14, 2010
    Exactly - we just have to see how this all pans out. I am eager to see what Intel has in their 18core mammoth chip :ROFLMAO:, of course some will be be saying it will hit 5ghz too :playful:. Still I think AMD needs to do much better on the platform end with Threadripper and Epyc.